No full-text available
To read the full-text of this research,
you can request a copy directly from the authors.
Support vector machine: A tool for mapping mineral prospectivity
Abstract
In this contribution, we describe an application of support vector machine (SVM), a supervised learning algorithm, to mineral prospectivity mapping. The free R package e1071 is used to construct a SVM with sigmoid kernel function to map prospectivity for Au deposits in western Meguma Terrain of Nova Scotia (Canada). The SVM classification accuracies of ‘deposit’ are 100%, and the SVM classification accuracies of the ‘non-deposit’ are greater than 85%. The SVM classifications of mineral prospectivity have 5–9% lower total errors, 13–14% higher false-positive errors and 25–30% lower false-negative errors compared to those of the WofE prediction. The prospective target areas predicted by both SVM and WofE reflect, nonetheless, controls of Au deposit occurrence in the study area by NE–SW trending anticlines and contact zones between Goldenville and Halifax Formations. The results of the study indicate the usefulness of SVM as a tool for predictive mapping of mineral prospectivity.
... A total of 118 tungsten occurrences, including historical mines, discovered deposits, and verified prospects, were employed as training samples. The study area was subdivided into 195,174 predictive units with a cell size of 450 m, following the rasterizing scenario of our previous work [18], based on the criterion proposed by Zuo and by Carranza [64,65]. The evidential features were transformed into raster maps in which each cell has a numerical representation of the features. ...
... Numerous methods and procedures are involved in this study, including data preparation, fractal and multifractal analyses, feature selection, AI model training, model assess- Figure 2 illustrates the flowchart of the proposed framework; the following sections describe in detail the key methods and processes used, i.e., fractal and multifractal methods, feature selection, and AI-driven models. [64,65]. The evidential features were transformed into raster maps in which each cell has a numerical representation of the features. ...
... The labelled samples were derived from our previous study, conducted in the same area [101], which comprised 118 known tungsten occurrences (positive samples) and 346 non-occurrences (negative samples). The negative samples were randomly selected according to the criteria proposed by Carranza and by Zuo [64,134]. Ten training datasets were then generated from the labelled dataset, and each training dataset included 118 positive samples and 118 negative samples that were randomly selected from 346 non-occurrences. ...
AI-driven mineral prospectivity mapping (MPM) is a valid and increasingly accepted tool for delineating the targets of mineral exploration, but it suffers from noisy and unrepresentative input features. In this study, a set of fractal and multifractal methods, including box-counting calculation, concentration–area fractal modeling, and multifractal analyses, were employed to excavate the underlying nonlinear mineralization-related information from geological features. Based on these methods, multiple feature selection criteria, namely prediction–area plot, K-means clustering, information gain, chi-square, and the Pearson correlation coefficient, were jointly applied to rank the relative importance of ore-related features and their fractal representations, so as to choose the optimal input feature dataset readily used for training predictive AI models. The results indicate that fault density, the multifractal spectrum width (∆α) of the Yanshanian intrusions, information dimension (D1) of magnetic anomalies, correlation dimension (D2) of iron-oxide alteration, and the D2 of argillic alteration serve as the most effective predictor features representative of the corresponding ore-controlling elements. The comparative results of the model assessment suggest that all the AI models trained by the fractal datasets outperform their counterparts trained by raw datasets, demonstrating a significant improvement in the predictive capability of fractal-trained AI models in terms of both classification accuracy and predictive efficiency. A Shapley additive explanation was employed to trace the contributions of these features and to explain the modeling results, which imply that fractal representations provide more discriminative and definitive feature values that enhance the cognitive capability of AI models trained by these data, thereby improving their predictive performance, especially for those indirect predictor features that show subtle correlations with mineralization in the raw dataset. In addition, fractal-trained models can benefit practical mineral exploration by outputting low-risk exploration targets that achieve higher capturing efficiency and by providing new mineralization clues extracted from remote sensing data. This study demonstrates that the fractal representations of geological features filtered by multi-criteria feature selection can provide a feasible and promising means of improving the predictive capability of AI-driven MPM.
... In recent years, machine learning (ML) methods have been used widely for MPM because they can effectively reveal complex and nonlinear spatial associations between known deposits and evidential features (Brown et al., 2000;Bergen et al., 2019;Zuo, 2020). Over the past decade, the state-of-theart data-driven methods of MPM have been comprised primarily of supervised machine learning and deep learning algorithms (Zuo and Carranza, 2011;Carranza and Laborte, 2015a;Rodriguez-Galiano et al., 2015;Xiong and Zuo, 2018;Zuo and Xu, 2023). Supervised learning methods, which have demonstrated potential in discovering specific geochemical patterns of interest, have been adopted widely in MPM since the 2010s. ...
... Supervised learning methods, which have demonstrated potential in discovering specific geochemical patterns of interest, have been adopted widely in MPM since the 2010s. These methods include artificial neural networks (ANN) (Porwal et al., 2003), logistic regression (LR) (Porwal et al., 2010), random forest (RF) Laborte, 2015a, 2015b), support vector machine (SVM) (Zuo and Carranza, 2011;Abedi et al., 2012) and extreme learning (Chen and Wu, 2017). More recent researches have introduced novel deep learning algorithms for MPM because they are highly capable of capturing mineralization information and the spatial interrelations between mineralization and prospecting data Zuo and Xu, 2023), such as convolutional autoencoder network (Zhang et al., 2021), deep convolutional neural network , deep GMDH (group method of data handling) neural networks , and graph deep learning (Zuo and Xu, 2023). ...
... Moreover, it can be used in data-driven MPM even when there are few training samples available (Carranza and Laborte 2015b). On the other hand, the SVM has also been widely applied in data-driven MPM and it is particularly suitable in cases of small sample sizes and large, high-dimensional datasets, as it requires only a small subset of data to construct the classifier (Cherkassky and Ma, 2004;Zuo and Carranza, 2011;Granek, 2016;Ge et al., 2022). ...
Although mineral prospectivity modeling (MPM) has undergone decades of development, it has not yet been widely adopted in the global mineral exploration industry. Exploration geoscientists encounter challenges in understanding the internal working of many mineral prospectivity models due to their black box nature. Besides, their predictive results usually delineate undesirably large high-prospectivity areas, which are biased toward existing deposits, making MPM impractical. However, there are only a few data-driven methods for MPM that address both the interpretability of black box models and the issue of bias in high prospective areas, which may result from the intrinsic stochastic uncertainty of training samples, particularly toward well-known deposits. In this study, we construct and demonstrate a framework to improve the performance and reliability of data-driven MPM in the Qulong–Jiama mineral district of Tibet. Firstly, the mineral systems concept was applied to select appropriate targeting criteria and to derive corresponding evidential features. Secondly, model-agnostic methods, such as permutation feature importance, partial dependence plot, individual conditional expectation plot, and Shapely values, were applied to interpret the machine learning models. Finally, modulated prediction models and the spatial pattern of linked uncertainties were generated by an ensemble method that combines bootstrapping and the Random Forest algorithm. The final exploration targets, which were demarcated by cells with high modulated values and low uncertainties obtained by 50 predictive models, account for just ~ 3% of the study area.
... GIS-based mineral prospectivity mapping (GMPM) is a convenient and effective method to obtain the prospective potential areas with low cost and short period (Carranza et al., 2008, Carranza, 2011McCuaig et al., 2010;Zuo and Carranza, 2011;Zuo, 2020a). The advantage of GMPM is integrating the multi-source mineralization information under the guidance of the mineral system conception model (Carranza et al., 2008;Ford et al., 2019), which greatly excludes the false abnormalities originating from a single prospective dataset and decreases the prospective potential areas. ...
... The advantage of GMPM is integrating the multi-source mineralization information under the guidance of the mineral system conception model (Carranza et al., 2008;Ford et al., 2019), which greatly excludes the false abnormalities originating from a single prospective dataset and decreases the prospective potential areas. Those methods are usually carried out by datadriven and knowledge-driven as well as hybrid spatial data models (Agterberg, 1989;Cheng and Agterberg, 1999;Porwal et al., 2003;McCuaig et al., 2010;Carranza, 2011;Zuo and Carranza, 2011;Carranza and Laborte, 2015;Wang and Zuo, 2022;Zuo and Xu, 2023). The datadriven utilizes the position of known deposits as training points and the knowledge-driven methods evaluate MPM based on expert judgment. ...
... Although the workflows of GMPM and GSMPM are similar, the latter can directly count and mine the geoscience data, or input the original data in support of MPM (Zuo et al., 2021). GSMPM or GMPM combined with machine learning algorithms had been applied in different deposit types to capture the complex relationships between multi-source geological data information and the known deposits (Porwal et al., 2003: McCuaig et al., 2010Carranza, 2011;Zuo and Carranza, 2011;Carranza and Laborte, 2015;Yousefi and Carranza, 2015;Yousefi and Nykänen, 2017;Zuo, 2020a), and verified the posterior possibility by receiver operating characteristic curve (ROC) (Zuo and Carranza, 2011;Carranza and Laborte, 2015;Chen, 2015;Xiong et al. 2018;Zhang et al., 2022). Different methods, including machine learning algorithms (e.g., neural networks, Brown et al., 2000;Singer and Kouda 1996, 1997, 1999, Porwal et al. 2003Radial Basis Functional Link Network (RBFLN) supervise and Fuzzy Clustering (FC) unsupervised machine learning methods, Looney and Yu, 2000;support vector machine, Zuo and Carranza, 2011;Rodriguez-Galiano et al., 2015; random forests (RF), Li et al., 2021a;Carranza and Laborte, 2015;extreme learning, Chen and Wu 2017) and deep learning algorithms (e.g., deep autoencoder network, Li et al., 2021b;Chen 2015; convolutional neural network (CNN), Li et al., 2020; geodata science-based mineral prospectivity mapping (GSMPM), Zuo and Wang, 2020 ;isolation forest, Zhang et al., 2022;Graph deep learning, Zuo and Xu, 2023) have been implied to promote the MPM. ...
The Zhunuo ore concentration area (ZOCA) is the most potential prospective area of Cu-Au (Mo) in the west of
the southern subterrane, Tibet. Single traditional prospective methods (e.g., stream sedimentary geochemistry)
often produced larger area and false abnormal information in the Gangdese orogenic belt because of the high
altitude and the intense weather and erosion, which can not meet the urgent demand of the current situation for
Cu resources. In this study, we combined a mineral system approach with GIS-based machine learning approachs to obtain geologically meaningful mineral prospective maps. The detail steps include: (i) establishing the
mineral system conception model of porphyry copper deposits (PCDs); (ii) transforming the targeted porphyry
metallogenic system components into spatial proxies associated with the crucial ore-forming processes; (iii)
extracting the spatial proxies: proximity to intrusive rocks (source), NE orientation faults (transport and/or
physical trap), Fe-oxide and propylitization hydrothermal alterations zone (hydrothermal fluids) and the metallogenic strength diagram of Cu-Mo-W-Bi-Au-Ag-Pb-Zn (deposition); (iv) Radial Basis Functions Link Networks
(RBFLN), Random forests (RF) Supervised and Fuzzy Clustering (FC) unsupervised machine learning methods
were applied to capture the complex and crucial mineralization information between known deposit types and
evidence layers; (vi) model estimation and delineating prospective potential targets: Receiver operating characteristic curve (ROC), predictive-area (P-A) plotting and normalised density (Nd) were used to evaluate the
predictive models results. The results indicate that the RBFLN model, RF model, and FC model show high predictive accuracy. The AUC values under the ROC area of the RBFLN model, RF model, and FC model are 0.99,
0.96, and 0.94, respectively. The RBFLN model outperforms the RF model and FC model, the predictive-area
plotting of RBFLN occupies 12% of the study area containing 88% of the known deposits. The predictive-area
plotting of the RF model and FC model showed that 14% and 21% of the study area contained 86% and 79%
of the known deposits, respectively. The normalized density (Nd) of a layer is defined as the ratio of the prediction success rate (Pr) of the P-A plotting to the corresponding area (Oa). The normalized density of the RBFLN
model, the RF model, and the FC model are 7.33, 6.14, and 3.76, respectively, which revealed that the results of
the three predictive models all have positive indications. These studies show that RBFLN supervised machine
learning method is a more robustness and generalization capability. The predictive results also provide prospective potential targets (e.g., northern Cimabanshuo, northwest Wubaduolai, and southwestern and western
Zhunuo PCD) for further exploration, and this method can be also applicable to other mineral systems and
districts.
... SVM models can be classified from simple to complex into linear 303 separable, linear non-separable, and nonlinear SVM algorithms. Linear SVM is an 304 interval-maximized search of the dividing line/hyperplane whereby the different classes 305 of sample points are distributed as evenly as possible on both sides of the dividing 306 line/hyperplane for accurate classification of the target (Burges, 1998;Cortes et al., 307 1995;Zuo et al., 2011). ...
... 329The original constrained optimization problem can be solved by transforming it 330 into a maximin dual problem according to the Lagrange dual problempenalty coefficients C are introduced to find the optimal separating hyperplane 334(Burges, 1998;Zuo et al., 2011), allowing the model to tolerate specific errors and 335 avoiding overfitting issues. The introduced usable objective function can be derived 11 in the equivalent dual problem of SVM the inner product of 339 sample points in the objective function is i j x x × . ...
The extraction and integrated analysis of multi-source geological data are key steps in the prediction of mineralization. Current studies are focusing on the extraction and integration of the deep-level mineralization information. In the era of big data, mathematical models and computer algorithms for data mining of multi-source prospecting information have emerged as a leading research area in mineral prediction. In this study, we quantitatively analyzed the structure and remote sensing alteration information using the concentration-area (C-A) fractal model and the box-counting method for the Duobaoshan mineralization area, Heilongjiang Province, China. Results indicate that areas of high fractal dimension of remote sensing alteration correspond to abundant alteration anomalies. Fractal characterization of geological structures is consistent with the spatial distribution. Therefore, fractal characterization provides predictive factors of structure and remote sensing alteration in the development of a predictive model of mineralization. Soil geochemical data were analyzed using the component data analysis (CDA) method and the spectrum-area (S-A) fractal model. The analyses identified anomalous and background signals represented by the PC1 and PC2 principal component combinations. These combinations show a strong correlation between geochemical anomaly data and known deposits in the study area, suggesting that the S-A model effectively identifies geochemical anomalies that can be used as a predictive factor of a mineralization prediction model. The mineralization prediction model was developed using random forest (RF) and support vector machine (SVM) algorithms. The model incorporates predictive factors from multiple sources, including the ore-forming geological background, fractal-characterized geological structure, fractal-characterized remote sensing alteration, and geochemical characteristics. The models incorporated the C-A fractal model to evaluate the probability of mineral prediction. By integrating the characteristics of multi-source mineral prospecting information with the predictive results of machine-learning models, we delineated eight prospective mineralization areas. This approach validates the effectiveness of a combined method involving fractal theory and machine-learning in mineral exploration, offering new insights and theoretical guidance for further mineral prospecting in the study area.
... However, the extension of data science into geodata science is more recent and ongoing (e.g., Zuo, 2020;Yousefi et al., 2021) (Fig. 1). Within MPM, and for strictly the purpose of data analysis, machine learning is used to primarily: (1) identify mineralization-related anomalies through unsupervised learning (e.g., Nwaila et al., 2022;Zhang et al., 2022b); (2) predict targets that are similar to known occurrences through supervised learning (e.g., Zuo & Carranza, 2011;Zhang et al., 2021;Senanayake et al., 2023); and (3) predict targets using reinforcement learning (e.g., Shi et al., 2023). Outside of data analysis, machine learning is also beginning to be used in: (1) data generation (e.g., Zhang et al., 2022b;Bourdeau et al., 2023); (2) data processing (e.g., Song et al., 2020;Nwaila et al., 2023;Zhang et al., 2023); and (3) simulations (e.g., Song et al., 2021). ...
... The range of all possible algorithms is unknowable because there are emerging algorithms and variations of existing ones, either as architectural modifications (e.g., changes in neural network architecture) or as add-ons (e.g., optimization algorithms; Chen et al., 2020;Yin & Li, 2022;Gharehchopogh et al., 2023). An empirical analysis revealed that algorithms used by various authors include (non-exhaustively): Bayes network (Porwal et al., 2006;Yin & Li, 2022); logistic regression (Agterberg & Bonham-Carter, 1999;Carranza & Hale, 2001;Karbalaei Ramezanali et al., 2020;Lin et al., 2020;Zhang et al., 2022c); support vector machines (Zuo & Carranza, 2011;Zhang et al., 2021;Senanayake et al., 2023); tree-based methods, such as random forest, extra trees and XGBoost (Chen & Wu, 2019;Sun et al., 2019;Zhang et al., 2022a); artificial neural networks, such as extreme learning machines (Chen & Wu, 2017); deep learning methods (Xiong et al., 2018;Wang et al., 2020;Yang et al., 2022;Zuo et al., 2022Li et al., 2023;Yin et al., 2023;Zuo & Xu, 2023); and reinforcement learning (Shi et al., 2023). There are also applicative MPM studies that employed ensemble learning, which is an approach to improve outcome reliability by integrating the output of multiple independent models (e.g., Senanayake et al., 2023;Shetty et al., 2023). ...
The primary goal of mineral prospectivity mapping (MPM) is to narrow the search for mineral resources by producing spatially selective maps. However, in the data-driven domain, MPM products vary depending on the workflow implemented. Although the data science framework is popular to guide the implementation of data-driven MPM tasks, and is intended to create objective and replicable workflows, this does not necessarily mean that maps derived from data science workflows are optimal in a spatial sense. In this study, we explore interactions between key components of a geodata science-based MPM workflow on the geospatial outcome, within the modeling stage by modulating: (1) feature space dimensionality, (2) the choice of machine learning algorithms, and (3) performance metrics that guide hyperparameter tuning. We specifically relate these variations in the data science workflow to the spatial selectivity of resulting maps using uncertainty propagation. Results demonstrate that typical geodata science-based MPM workflows contain substantial local minima, as it is highly probable for an arbitrary combination of workflow choices to produce highly discriminating models. In addition, variable domain metrics, which are key to guide the iterative implementation of the data science framework, exhibit inconsistent relationships with spatial selectivity. We refer to this class of uncertainty as workflow-induced uncertainty. Consequently, we propose that the canonical concept of scientific consensus from the greater experimental science framework should be adhered to, in order to quantify and mitigate against workflow-induced uncertainty as part of data-driven experimentation. Scientific consensus stipulates that the degree of consensus of experimental outcomes is the determinant in the reliability of findings. Indeed, we demonstrate that consensus through purposeful modulations of components of a data-driven MPM workflow is an effective method to understand and quantify workflow-induced uncertainty on MPM products. In other words, enlarging the search space for workflow design and experimenting with workflow components can result in more meaningful reductions in the physical search space for mineral resources.
... We integrated the layers shown in Fig. 2 using support vector machine, random forest, and extreme learning neural network to utilize the positive sides of supervised machine learning methods and training point sets. These supervised methods were trained using 22 Cu occurrences as positive samples and 22 non-deposit locations as negative samples, where mineral deposits are predicted as unlikely to be present (e.g., Zuo and Carranza, 2011;Zhang et al., 2016;Zuo and Wang, 2020;Rahimi et al., 2021). The negative point samples were randomly selected out of polygon and line features, which were applied as exploration evidence in this study. ...
... Unsupervised methods (Xiong et al., 2018;Zuo et al. 2019;Rahimi et al., 2021) reveal patterns among multiple exploration data, yet may not produce results consistent with known deposit locations and knowledge. Supervised machine learning approaches are usually consistent with known deposit locations and exploration knowledge as they use positive and negative training point sets to constrain spatial predictions (e.g., Zuo and Carranza, 2011;Carranza, and Laborte, 2016;Zhang et al., 2016;Sun et al., 2019;Xiong and Zuo, 2021). The geometric average integration approach (Yousefi and Carranza, 2015b) introduces exploration targets in a way that respects uncertainty in the input prospectivity values among an exploration dataset. ...
... Several fields of activity are covered (biomedical [3], facial recognition [4] and fingerprints, etc.). Machine learning has been successfully applied to the analysis of natural phenomena such as potential earthquakes [5], volcanic eruptions [6], classification of seabed mud volcanoes [7] and mining prospects [8]. However, only a few research projects using computer vision and image processing are concerned with rock classification [1] [9] [10] [11] [12] [13]. ...
... The experimental study was conducted with a data set comprising eight (8) classes of direct-view digital images of magmatic (granite, granodiorite, gabbro) and metamorphic (schist, cipolin, migmatite, eclogite, hornfels) rocks, totalling eighty (80) images. This grouping into eight (8) classes was made by an expert geologist, taking into account certain objective characteristics of textures and colors. ...
... Recently, machine learning, in particular for deep learning, with powerful ability of data transfer and abstraction and high-level feature extraction have been increasingly applied in MPM (Zuo et al., 2019Lawley et al., 2021;Xiong and Zuo, 2021), such as random forest (Xiang et al., 2020;Talebi et al., 2022), support vector machine (Zuo and Carranza, 2011;Mao et al., 2019;Prado et al., 2020), artificial neural networks (Diaz-Rodriguez et al., 2021), convolutional neural network (CNN; Li et al., 2023b;Ding et al., 2023), gated recurrent unit model , long short-term memory networks (Wang and Zuo, 2022), tensor dictionary learning (Yu et al., 2022), variational autoencoder , deep self-attention model (Yin et al., 2023), and graph deep learning model Zuo and Xu, 2023). Generally, shallow machine learning using pixel/voxel data is applied more frequently in 3D MPM (e.g., Yuan et al., 2014;Mao et al., 2019Mao et al., , 2020Fu et al., 2021). ...
... In this case, the input data (total 580 GB) for model training and prospectivity consisted of five features (dF, gF, vF, waF and wbF) of the Sanshandao fault model (surface to -3000 m). Since negative samples are critical for the prospectivity modeling (Zuo and Carranza, 2011;Prado et al., 2020), we primarily utilized voxels verified by drillholes near known orebodies to maintain a similar distribution range of several variables (Supplementary Fig. 1). The well-known regions dataset comprises 34,460 voxels, divided based on an Au grade greater than 1 g/t into 15,356 mineralized voxels and 19,104 non-mineralized voxels. ...
... The negative samples can be obtained either by randomly selecting the location of geological constraint points or by selecting barren drilling locations (Carranza et al, 2008;Nykä nen et al, 2015;Zuo et al., 2021). The supervised learning methods include logistic regression (Xiong and Zuo, 2018;Wang et al., 2021a;Zhang et al., 2022), random forests (Carranza and Laborte, 2015;Gao et al., 2023a, b), support vector machine (SVM) (Zuo and Carranza, 2011), neural networks (Porwal et al., 2003;Li et al., 2021), XGBoost (Chen and Gusetrin, 2016). The semi-supervised learning method is the semi-supervised random forest (Wang et al., 2020a). ...
... The main concept of SVM is to form an edge classifier with optimum complexity according to the number of support vectors, which is appropriate for small samples and large high-dimensional datasets (Granek, 2016;Ge et al., 2022). The SVM is widely used in binary classification tasks (Cortes and Vapnik, 1995), including MPM (Zuo and Carranza, 2011;Ge et al., 2022). The OCSVM and SVM have in common that they separate complex nonlinear data, introduce kernel function to convert low-dimensional data into high-dimensional feature space, and use the optimum hyperplane in the high-dimensional feature space to separate data in the feature space with the maximum interval (Cortes and Vapnik, 1995;Schö lkopf et al., 2001;Chen and Wu, 2017;Ge et al., 2022). ...
Three-dimensional (3D) mineral prospectivity mapping (MPM) uses mathematical models to integrate different types of 3D data related to mineralization to obtain mineral prospectivity information in 3D space. Existing geological data contain known deposits, non-deposits and unknown ore-bearing data, corresponding to positive samples, negative samples and unlabeled samples respectively in MPM. Different sample combination types require different mathematical models. In this paper, support vector machine class (SVMC) machine learning method is selected to compare the influence of different sample combination types on prediction results. The SVMC is a one-class SVM (OCSVM) model based on positive-only samples, the SVM is based on both positive and negative samples, and the bagging-based positive-unlabeled learning algorithm with SVM base learner (BPUL-SVM) is based on both positive and unlabeled samples. The study area is in the Sanshandao-Cangshang offshore and onshore Au district, where there are Sanshandao, Cangshang and Xinli large- and super-large-scale Au deposits. Moreover, the discovery of large-scale Sea Au deposits in the sea area indicates the great potential for mineralization in the district. According to the metallogenic geological characteristics, the Au deposits in the Sanshandao-Cangshang district are controlled by the NE-striking fault and are closely related to the Linglong intrusions and Guojialing intrusions. The ore-bearing intrusion shows low density and low-moderate magnetic susceptibility. Because the Au orebodies hosted in the Sanshandao fault and its secondary faults, the NE-striking faults are key to delineating the targets. In this paper, weights of evidence (WofE), OCSVM, SVM and BPUL-SVM are used to MPM, and the prediction-area (P-A) plot method is used to delineate the targets. According to the ROC curve, F1 score and P-A plot evaluation methods, the model performance from high to low is BPUL-SVM13, SVM12, WofE and OCSVM. The BPUL-SVM model performance with samples combination types of positive samples and unlabeled samples was optimum in SVMC prediction models. The Markov chain Monte Carlo (MCMC) simulation and return-risk evaluation model are used to evaluate the return and risk of the targets and finally determine the I-level targets with high return and low risk. The delineated targets are mainly distributed along the F2 and F3 faults (Sanshandao-Cangshang fault). Combined with the mineralization regularity, the deep and periphery space of the known deposits are important to explore Au orebodies. The delineated targets are important to explore offshore and onshore Au orebodies in the Sanshandao-Cangshang district.
... The support vector machines (SVM) method [7,107] consists of a set of supervised learning algorithms based on the statistical learning theory [108]. The dataset used is labeled with known class labels, and the algorithm is trained to project an ideal linear hyperplane that optimizes the distance between the two closest sample points (support vectors) to separate multiple classes. ...
... Non-linear datasets are converted into linear ones by transferring input data to a higher-dimensional feature space using kernel functions. In this work, the radial basis function (rbf) kernel function was selected because it has lower error rates compared to others [107], and it requires simple parameters for geoscientific data applications, namely the parameter C (balancing errors, margin width, and the number of support vectors) and the parameter γ (related to the width of the distribution and optimized for better results). ...
This work aims to model mineral prospectivity for intrusion–related gold deposits in the central portion of the Tapajós Mineral Province (TMP), southwestern Pará state. The scope includes experimentation and evaluation of knowledge and data-driven methods applied to multisource data to predict potential targets for gold mineralization. The radiometric data processing allowed to identify a hydrothermal alteration footprint of known gold deposits, providing information in regions with little or no field data available. The aeromagnetic data analysis prompted the identification of high magnetic zones, which are probably related to hydrothermal fluid transport. Linear features extracted from digital elevation data revealed an NNW–SSE general trend, which is consistent with the main structural control of deposits. The data were integrated through three modeling techniques—fuzzy logic (knowledge-driven), weights of evidence (WofE, data-driven), and a machine learning algorithm (SVM, data-driven)—resulting in three prospective models. In all models, the majority of indicated prospective regions coincide with the known deposits. The results obtained in the models were combined to generate an agreement map, which mapped the overlapping of their highest prospective scores, indicating new areas of prospective interest in the central portion of the TMP.
... The application of machine learning (ML) and artificial intelligence (AI) techniques has signifi-cantly enhanced mineral prospectivity mapping (Rodriguez-Galiano et al., 2015). ML algorithms, such as random forests (Carranza & Laborte, 2015;Carranza & Laborte, 2016), support vector machines (Zuo & Carranza, 2011) and neural networks (Porwal et al., 2003b;Nykä nen, 2008), can effectively analyze large datasets and identify complex patterns. These methods enable the extraction of valuable information from geological and geospatial data, leading to accurate mineral potential predictions and the identification of previously unknown mineralization targets. ...
... Based on Bonham-Carter (1994), mineral prospectivity mapping methodologies have two dominant end members: empirical modeling methods (data-driven approach) and conceptual modeling methods (knowledge-driven approach). The first category includes procedures that require known mineral deposits that are used as prior knowledge to train the models (e.g., Bonham-Carter et al., 1989;Bonham-Carter, 1994;Pan & Harris, 2000;Carranza & Hale, 2001a;Mihalasky & Bonham-Carter, 2001;Carranza & Hale, 2002;Harris et al., 2003;Porwal et al. 2003aPorwal et al. , 2003bCarranza et al., 2005;Carranza, 2008;Nykä nen, 2008;Nykä nen et al., 2008b;Carranza, 2009;Zuo & Carranza, 2011;Abedi & Norouzi, 2012;Carranza & Laborte, 2016;Zuo & Wang, 2020). Empirical models are especially suitable for areas with large amounts of data, such as brownfield exploration terrains, that have many previously recognized mineral occurrences or deposits to be used for training the models and gaining information about the mineral systems (Yousefi et al., 2021). ...
This paper describes mineral prospectivity research conducted in Finland to predict favorable areas for cobalt exploration using the “fuzzy logic overlay” method in a GIS platform and public geodata of the Geological Survey of Finland. Cobalt occurs infrequently as a core product in mineral deposits. Therefore, we decided to construct separate conceptual mineral prospectivity models within the Northern Fennoscandian Shield, Finland, for four deposit types: (1) “ Orthomagmatic Ni–Cu–Co sulfide deposits, ” (2) “ Outokumpu-type mantle peridotite-associated volcanogenic massive sulfide (VMS)-style Cu–Co–Zn–Ni–Ag–Au deposits, ” (3) “ Talvivaara black shale-hosted Ni–Zn–Cu–Co-type deposits” and (4) “Kuusamo-type (orogenic gold with atypical metal association) Au–Co–Cu–U–LREE deposits ”. In addition, we created a model combining till geochemical data with data derived from bedrock drilling and mineral indications, including boulders and outcrops. The mineral prospectivity models were statistically tested with the “ receiver operating characteristics ” method using exploration drilling data from known mineral deposits as validation sites. In addition, the predictive performance of the models was evaluated by using success rate curves, where the number of previously identified deposits was compared with the area coverage of the predicted highly favorable areas. These results indicate that the knowledge-driven mineral prospectivity method using parameters derived from mineral systems models is effective in defining favorable exploration target areas at the regional scale. This study's innovation lies in its comprehension of the process of evaluating mineral prospectivity when the commodity of interest is not the primary commodity within the mineral system.
... The established relationships are used to rank the importance of each evidential layer relative to one another including the assignment of weights to the evidential layers for the generation of a reliable predictive mineral potential map of the mineral deposit sought (Chen et al. 2018;Forson et al. 2022). The popular data-driven models developed for use are random forest (RF), neural network (NN), support vector machine (SVM), prediction-area (P-A) plot, clustering methods, evidential belief functions (EBF) etc. (Porwal et al. 2006;Zuo and Carranza 2011;Abedi and Norouzi 2012;Yousefi and Carranza 2015;Chen et al. 2018;Sun et al. 2020;Forson et al. 2022). However, the major weakness of the datadriven models is the availability of enough exploratory data to establish a spatial relationship between mineral occurrences and geospatial anomalies before a reliable predictive mineral potential map can be produced (Chen et al. 2018). ...
In this paper, the knowledge-driven fuzzy AHP (FAHP) model was applied in the predictive prospectivity mapping of
orogenic gold deposits using the mineral system approach. The main criteria of the mineral system of orogenic gold miner-
alization considered were heat source, gold/ligand source, structural control, and hydrothermal alteration. The proxies and
alternatives of the main criteria of the mineral system of orogenic gold mineralization were derived from geological, geo-
physical, and remote sensing datasets. Assignment of weights using the FAHP model indicates that heat source with weight
of 0.4318 has the highest contribution to orogenic gold mineralization in the study area. This was closely followed by gold/
ligand source (0.3669), structural control (0.1659) and hydrothermal alteration (0.0355). The integration of these criteria
using the multi-index overlay method produced the predictive mineral potential map (MPM) which was further classified
into six classes (background, very low, low, moderate, high, and very high potentials) using the concentration area (C–A)
fractal model. New major prospects of orogenic gold mineralization were delineated in the western and eastern flanks of the
study area. The validation of the produced MPM using 10 geochemical sampling points via the prediction area (P–A) plot
yielded 77% success rate indicating the model is suitable for predictive prospecting of orogenic gold mineralization. The
study concluded that the mineral system approach should be adopted in further research in similar geologic environments
for reliable potential mapping of any mineral deposits because it considers every “player” involved in mineralization.
... Every instance has a type named y i , that either equals 1 for a particular group or À1 for the alternative group, depending on the scenario at hand (i.e., y i 2 f À 1, 1gÞ (Huang et al., 2002). If the categories can be divided linearly, splitting hyperplanes-a set of linear separatorsoccur (Zuo & Carranza, 2011). Support vector machines (SVMs) were used to search the "training" dataset for similar characteristics with the goal to reach this determination. ...
Environmental innovation (EI) is fundamental to sustainable development goal (SDG) number 9. Indirectly, it contributes to the achievement of SDG 7 by laying the groundwork for producing renewable energy. Firms involve environmental, social, and governance (ESG) and diversity practices to achieve sustainable success. ESG and diversity scores on EI need to be predicted, yet EI predictors are few in the research. Our institutional theory-based study examines whether ESG and diversity scores influence EI scores in multinational organizations. The dataset comprises information from the Refinitiv Eikon database, including 430 publicly traded firms worldwide throughout 2021. The results of our study indicate that the environmental pillar score, ESG, and workforce score are the three most significant factors for calculating enterprises’ EI scores. This research provides valuable insights into enhancing sustainability practices and fostering innovation in global firms, offering a practical roadmap for businesses striving to achieve these objectives.
... Data-driven models for mineral prediction are estimated based on quantitative measures of spatial associations between evidential characteristics and known deposits of the targeted type (Carranza 2011;Carranza and Laborte 2015;Chen and Wu 2017;Fatehi and Asadi 2017;Karpatne et al. 2018;Sun et al. 2019Sun et al. , 2020Qin et al. 2021;Parsa et al. 2021;Lin et al. 2021;Yousefi et al. 2021;Zuo and Xu 2022;Hajihosseinlou et al. 2024). Such as logistic regression (Xiong and Zuo 2018;Zhang et al. 2018;Xiao et al. 2020Xiao et al. , 2022, weights of evidence (Zhang and Zhou 2015;Zhang et al. 2016;Xiao et al. 2018Xiao et al. , 2020Xiao et al. , 2023Fu et al. 2021;Tao et al. 2021;Zhang et al. 2022), Bayesian network classifiers (Porwal et al. 2006), neural networks (Rigol-Sanchez et al. 2003;Oh and Lee 2010;Tayebi et al. 2014;Rodriguez-Galiano et al. 2015;Saljoughi and Hezarkhani 2016;Maepa et al. 2021;Chen et al. 2022), support vector machine (SVM) (Zuo and Carranza 2011;Abedi et al. 2012;Rodriguez-Galiano et al. 2015;Saljoughi and Hezarkhani 2016;Ke et al. 2018;Xu et al. 2020;Maepa et al. 2021;Ghezelbash et al. 2021), radial basis functional link nets (Nykänen 2008), random forest (RF) (Carranza and Laborte 2015;Rodriguez-Galiano et al. 2015;Zhang et al. 2016;Ford 2020;Xiang et al. 2020;Daviran et al. 2021), extreme learning machines (Chen and Wu 2017), restricted Boltzmann machines (Chen et al. 2014), decision trees (Rodriguez-Galiano et al. 2015), and maximum entropy models (Liu et al. 2018). These methods use mathematical models to establish quantitative associations between known deposit types and prospecting indicators. ...
The world has moved into an era of hidden ore body exploration, necessitating the development of new prospecting and exploration methods. One promising approach is to use the deep convolutional neural network (DCNN) algorithm to extract spatial and correlation characteristics of multiple two-dimensional elements related to hidden ores. This paper explores this method on Datangpo manganese (Mn), constructing prediction datasets that includes geological, geochemical, geophysical and aeromagnetic features. Analyzing metallogenic conditions and control factors of Mn ores, we construct a Mn ore prediction model (Geo-DCNN) based on multiple geographical knowledge and DCNN. The Geo-DCNN model reaches ore-bearing accuracy of 79.11%, non-ore-bearing accuracy of 99.01%, overall accuracy of 95.35%, and loss value of 0.0227 after training. Based on analysis of ROC curve, P-R curve, field investigation, and target area verification, we discover that the prediction results of the Geo-DCNN model in northeast Guizhou have a high correspondence rate with known manganese deposits. This provides valuable insight for further ore exploration in the area. Additionally, the results indicate that the Geo-DCNN model is robust and portable, suggesting that it can be applied to metallogenic prediction practices for manganese ore in similar regions.
... It creates a set of hyperplanes to distinguish between instances of different classes and seeks to find the optimal hyperplane that maximizes the margin between the different classes (Xie et al., 2018;Bern et al., 2021;Chen et al., 2023). It has been applied to map prospectivity for gold deposits in Nova Scotia, Canada (Zuo and Carranza, 2011), classify mudstone lithofacies in the complex depositional environments of the Bakken and Marcellus shales of North America (Bhattacharya et al., 2016), and detect gold mineralization-related geochemical anomalies in Hebei Province, China (Chen et al., 2023). ...
Rare earth elements (REE), classified as critical minerals which are crucial for clean energy technologies, face soaring demand. While economic deposits are found in limited geologic environments including carbonatites and ion-adsorption clays, unconventional, secondary sources such as those from sedimentary basins could hold potential to meet this increased demand. Coal and its associated combustion by-products, phosphorites, oil sands tailings, and formation waters have all garnered interest for REE recovery, yet they remain significantly underexplored. Accordingly, new tools for data analysis and optimization such as machine learning can assist in mineral prospectivity, with these tools being subject to rapid proliferation in the Earth sciences.
This work leverages compositional data analysis principles and machine learning to probe geochemical relationships and predict REE abundances in sedimentary lithologies using unsupervised (correlation, principal component, and cluster analysis) and supervised (regression, support vector machine, random forest, and boosting) machine learning models. These three unsupervised models display similar results, with REE typically being associated with incompatible elements (e.g., Th, Nb, and Hf). Gradient boosting, Adaboost, and Random Forest had the highest performance for predicting REE concentrations, with Th and P commonly being the most important predictor variables. Identifying geochemical indicators of REE enrichment that may be used to assist in discovering potentially exploitable REE resources based on existing data, as well as increasing the understanding
of metal behaviour in sedimentary systems, is a step forward in understanding novel secondary and unconventional REE sources. Although REE concentrations from these sources are generally lower than primary ore deposits, the amount of available feedstock, potentially simpler, cheaper, and less environmentally taxing extraction processes, and the added benefit of remediating waste streams and contributing to the circular economy make these sources alluring.
... By considering expert knowledge of or quantifying spatial association of known training data for weighting and ranking geospatial proxies in mineral exploration, data integration modeling can be classified, respectively, into knowledge-driven (e.g., Barak et al., 2021;Yousefi and Carranza, 2015;McCuaig et al., 2010;Carranza and Hale, 2002) or datadriven (e.g., Zhang et al., 2022;Carranza and Laborte, 2015;Zuo and Carranza, 2011) methods. A third category is hybrid data integration modeling, whereby both expert knowledge of and quantified spatial association of known training data are used to tune the weights of spatial proxies for mineral prospectivity mapping "MPM" (e.g., Yang et al., 2022;Parsa et al., 2017;Asadi et al., 2015). ...
... Mineral prospectivity mapping (MPM) aims to delineate potential regions for discovering new mineral deposits by integrating diverse prospecting information, and it has vital value for mineral exploration (Bonham-Carter, 1994;Agterberg, 2021;. Artificial intelligence (AI) algorithms, as extraordinary data-driven MPM models, such as shallow machine learning (ML), deep learning (DL), and interactive reinforcement learning, have been reported extensively in the literature due to their exceptional ability to extract nonlinear features (Porwal et al., 2006;Zuo & Carranza, 2011;Rodriguez-Galiano et al., 2014;Zuo, 2020;Li et al., 2021a;Singer, 2021;Yin et al., 2022;Shi et al., 2023a;, 2024. Among numerous AI algorithms, random forests (RFs) and convolutional neural networks (CNNs), representative ML and DL algorithms, respectively, have gained significant recognition from geoscientists owing to their prominent advantages (LeCun et al., 2015;Parsa et al., 2018;Xiong et al., 2018;Li et al., 2020Li et al., , 2021bYang et al., 2022Yang et al., , 2023Zuo & Carranza, 2023). ...
Various artificial intelligence (AI) algorithms have been employed successfully to map mineral prospectivity for a specific mineral deposit type to assist mineral exploration. Numerous tools have been developed to incorporate AI algorithms, such as ArcSDM and ArcGIS. However, existing tools remain inadequate for geologist-friendly functions, and they are not fully tailored for mineral prospectivity mapping (MPM). This limitation has impeded the advancement and utilization of AI algorithms in MPM. Thus, this study introduced a novel ArcEngine-based software named ArcMPM to expeditiously integrate multi-source prospecting information for MPM using AI algorithms. ArcMPM was developed using Python and C#, based on ArcEngine and Visual Studio 2012, which incorporate two popular machine learning (ML) approaches: random forests (RFs) and convolutional neural networks (CNNs), representing shallow ML and deep learning algorithms, respectively. Moreover, it encompasses a complete procedure suitable for MPM by utilizing the RF and CNN models from sample generation to model evaluation. A case study in the Baguio region of the Philippines illustrated the convenience and effectiveness of utilizing ArcMPM for MPM. The success-rate curves demonstrated that the RF and CNN models developed in ArcMPM, particularly the CNN, exhibited high accuracy in delineating high-prospectivity areas. In addition, the case study proved that, in contrast to other GIS tools, ArcMPM can conveniently generate positive and negative samples under geological constraints, customize the model structure to suit the MPM according to the needs of geologists, and provide evaluation metrics that are accessible and practical to geologists.
... Applied Geochemistry 160 (2024) 105857 of mineral systems, machine learning methods have become useable a powerful data-fusion tool for integrating multi-source geoinformation, including both explicit and implicit characteristics for predictive mapping of mineral prospectivity (Porwal et al., 2010Sun et al., 2019;Xiao et al., 2020;Li et al., 2021). Examples of machine learning methods include logistic regression (Agterberg, 1974;, support vector machine (Zuo and Carranza, 2011;Abedi et al., 2012), artificial neural networks (Porwal et al., 2003;Ghezelbash et al., 2020), random forests (RF; Rodriguez-Galiano et al., 2015;Carranza and Laborte, 2015;Harris et al., 2015), and deep learning algorithms (Deng et al., 2020;McMillan et al., 2021;Zhang et al., 2021;Yin et al., 2022;Parsa et al., 2022;Yu et al., 2022;Zuo and Xu, 2023). Among these methods, RF is a robust and ensemble data-driven method that aggregates a collection of decision trees by randomly sampling the input dataset and variables, thus producing the final model by a majority vote on all trees (Rodriguez-Galiano et al., 2015;Carranza and Laborte, 2015;Harris et al., 2015). ...
... In recent decades, due to the uplifting in demand for base metals and also the decreasing trend of surface/shallow ore reserves, geoscientists have endeavored to trace mineralized zones in different geological settings through applying complex mathematical functions in the framework of the mineral prospectivity mapping (MPM) (Porwal et al. 2006;Carranza 2008;Zuo and Carranza 2011;Zuo and Xiong 2018;Chen and Wu 2017;Ghezelbash et al. 2021;Ghezelbash et al. 2020aGhezelbash et al. , 2023bDaviran et al. 2021Daviran et al. , 2023b. MPM is a multi-step technique that is able to process and use various geo-exploration datasets (e.g., geology, geochemistry, geophysics and hydrothermal alterations) to delineate the promising areas at preliminary to detailed exploration scales (Carranza 2008;Porwal and Kreuzer 2010;Harris et al. 2015a, b;Roshanravan et al. 2020). ...
Precisely selecting the exploration criteria and building robust machine-learning models are two critical issues for enhancing the efficiency of mineral prospectivity mapping (MPM) for delimiting highly favorable mineralization zones. The efficient exploration features linked to geochemical, geological, and remote sensing criteria were distinguished in the Torud-Chahshirin (TCS) volcano-intrusive belt, NE Iran using success-rate curves. Then, the Extreme Gradient Boosting (XGBoost) as an ensemble learning method was employed on a coherent group of exploratory evidence layers for highlighting the epithermal-Cu prospectivity areas in the TCS belt. In the next step, the artificial neural networks (here, MLP-ANN) as a data-driven machine learning technique was applied to compare the results which was obtained by the XGBoost algorithm. The outcomes of the receiver operating characteristics (ROC) curves illustrate that both predictive models succeeded in delineating target zones. However, regarding the area under the curve (AUC) values, the XGBoost model successfully delineates the exploration target by mostly Cu mineral occurrences rather than the MLP-ANN model.
... A Comparison of the integration results demonstrated that the RF is superior to GA for targeting areas with potential for porphyry-Cu deposits. It is important to note here that, RF unlike non-parametric supervised classifiers known as single classifiers, such as support vector machines [5], [87], deliver excellent results in the field of MPM. This is because the RF algorithm is not sensitive to the quality of training samples and overfitting. ...
Geochemical exploration data play a vital role in mineral prospectivity modeling (MPM) for discovering unknown mineral deposits. In this study, the improved spatially weighted singularity mapping (SWSM) method is used to improve the practice of identifying geochemical anomalies related to copper mineralization in the Sarduiyeh district, Iran. Then, the random forest algorithm (RF) and geometric average function (GA) are used to integrate the resulting geochemical predictor map with other predictor maps. As demonstrated by the high area under the curve (AUC) values, this approach can effectively delineate prospective areas with RF and GA. However, compared to the GA approach (AUC=0.78), the RF technique (AUC=0.98) offers superior prediction capabilities due to its enhanced ability to capture the spatial correlations between predictive maps and known mineral deposits. The proposed procedure, a hybrid of the improved SWSM, RF has outstanding predictive capabilities with concerning for identifying prospective areas. A case in point is the new, highly prospective areas identified in this study, which present priority targets for future exploration in the Sarduiyeh district.
... In recent years, many machine learning and deep learning methods have been developed for data-driven modeling (Lewkowski, Porwal, & Gonza'lez-A'lvarez, 2010). Decision tree (Breiman, 2017;Elith, Leathwick, & Hastie, 2008), artificial neural network (ANN) (Brown, Gedeon, Groves, & Barnes, 2000;Porwal, Carranza, & Hale, 2003), support vector machine (SVM) (Zuo & Carranza, 2011;Abedi, Norouzi, & Bahroudi, 2012), random forest (RF) (Breiman, 2001;Rodriguez-Galiano, Chica-Olmo, & Chica-Rivas, 2014), etc., are widely used in these methods. ...
... Mineral Prospectivity Mapping (MPM) is a geoscientific process that involves assessing and predicting the likelihood of discovering economically viable mineral deposits in a given known deposits, other types of known deposits, or random locations [21,47]. However, the study of Zuo and Carranza [18] shows that the selection of negative samples can affect the performance of MPM. At the same time, not all non-metallogenic locations are selected as negative samples, which means that the creation of negative sample datasets introduces uncertainty. ...
Three-dimensional Mineral Prospectivity Mapping (3DMPM) is an innovative approach to mineral exploration that combines multiple geological data sources to create a three-dimensional (3D) model of a mineral deposit. It provides an accurate representation of the subsurface that can be used to identify areas with mineral potential. These 3D geological models are the typical data source for 3D prospective modeling. Geological data sets from multiple sources are used to construct 3D geological models. Since in practice there is a significant imbalance in the ratio of mineralized to non-mineralized classes, the classification results will be biased in favor of the more observed classes. Borderline-SMOTE (BLSMOTE) is an oversampling technique used to solve the problem of unbalanced datasets and works by generating synthetic data points along the boundary line between the minority and majority classes. This helps to create a more balanced dataset without introducing too much noise. Non-mineralized samples can be generated by randomly selecting non-mineralized locations, which means that uncertainties are generated. In this paper, we take the shallow-forming low-temperature hydrothermal deposit Guizhou Lannigou gold deposit as an example to extract the ore-controlling elements and establish a 3D geological model. A total of 50 training samples are generated using the sampling method described above, and 50 mineralization prospects are generated using Random Forests. A return–risk analysis was used to explore the uncertainties associated with synthetic positive samples and randomly selected negative samples, and to determine the final mineral potential values. Based on the evaluation metrics G-mean and F-value, the model using BLSMOTE outperforms the model without the synthetic algorithm and the models using SMOTE and KMeansSMOTE. The optimal model BLSMOTE18 has an AUC of 0.9288. The methodology also performs superiorly with different levels of class imbalance datasets. Excluding the predictions where the results highly overlap with known deposits, five target zones were circled for the targets using a P-A plot, all of which have obvious metallogenic geological features. Among them, Target1 and Target2 have good potential for mineralization, which is of great significance for future mineral exploration work.
... With the development of big data and artificial intelligence technologies, machine learning algorithms have increasingly been applied to the field of geochemistry, providing a new perspective for the interpretation of geochemical data. Previous studies have highlighted various methods such as random forests [25,26], support vector machines [27,28], one-class support vector machines [29][30][31], neural networks [32,33], metric learning [34], and deep learning [35][36][37][38][39][40][41]. These studies have demonstrated the tremendous potential of artificial intelligence technologies, accelerating new discoveries and innovative solutions in the field of geochemistry. ...
With the rapid development of modern geochemical analysis techniques, massive volumes of data are being generated from various sources and forms, and geochemical data acquisition and analysis have become important tools for studying geochemical processes and environmental changes. However, geochemical data have high-dimensional, nonlinear characteristics, and traditional geo-chemical data analysis methods have struggled to meet the demands of modern science. Nowadays, the development of big data and artificial intelligence technologies has provided new ideas and methods for geochemical data analysis. However, geochemical research involves numerous fields such as petrology, ore deposit, mineralogy, and others, each with its specific research methods and objectives, making it difficult to strike a balance between depth and breadth of investigation. Additionally, due to limitations in data sources and collection methods, existing studies often focus on a specific discipline or issue, lacking a comprehensive understanding of the bigger picture and foresight for the future. To assist geochemists in identifying research hotspots in the field and exploring solutions to the aforementioned issues, this article comprehensively reviews related studies in recent years, elaborates on the necessity and challenges of combining geochemistry and artificial intelligence, and analyzes the characteristics and research hotspots of the global collaboration network in this field. The study reveals that the investigation into artificial intelligence techniques to address geochemical issues is progressing swiftly. Joint research papers serve as the primary means of contact within a worldwide collaborative network. The primary areas of focus in the ongoing research on the integration of geochemistry and artificial intelligence include methodologies for analyzing geochemical data, environmental modifications, and mineral prospectivity mapping. Geochemical data analysis is currently a significant focus of research, encompassing a range of methods including machine learning and deep learning. Predicting mineral resources for deep space, deep Earth, and deep sea is also a pressing topic in contemporary research. This paper explores the factors driving research interest and future trends, identifies current research challenges, and considers opportunities for future research.
... Meanwhile, knowledge-driven techniques require the geoscientist to weigh each predictor map with a subjective weight; the predictor maps are then summed to produce an MPM map. Many types of data-driven algorithms exist, including weights of evidence [1,2,[6][7][8][9][10][11], logistic regression [6,10,12,13], evidential belief modeling [14,15], support vector machine (SVM) [16,17], neural networks [18][19][20][21][22][23][24][25], and random forest, among others. Knowledge-driven techniques comprise algorithms such as Boolean, index overlay, and fuzzy logic [1,2,[26][27][28][29]. ...
Random Forest classification was applied to create mineral prospectivity maps (MPM) for orogenic gold in the Rainy River area of Ontario, Canada. Geological and geophysical data were used to create 36 predictive maps as RF algorithm input. Eighty-three (83) orogenic gold prospects/occurrences were used to train the classifier, and 33 occurrences were used to validate the model. The non-Au (negative) points were randomly selected with or without spatial restriction. The prospectivity mapping results show high performance for the training and test data in area-frequency curves. The F1 accuracy is high and moderate when assessed with the training and test data, respectively. The mean decrease accuracy was applied to calculate the variable importance. Density, proximity to lithological contacts, mafic to intermediate volcanics, analytic signal, and proximity to the Cameron-Pipestone deformation zone exhibit the highest variable importance in both models. The main difference between the models is in the uncertainty maps, in which the high-potential areas show lower uncertainty in the maps created with spatial restriction when selecting the negative points.
... There are several approaches to creating a set of negative samples (Qi et al., 2005;Zuo and Carranza, 2011;Butterworth et al., 2016), each with trade-offs for over-training/over-fitting, classification accuracy, but most importantly, for maximising the predictive power of the model. Negative samples generated randomly run the risk of choosing regions that may hold unexplored economic resources. ...
In recent years, the pace of technological development has accelerated along with the demand for minerals critical to sectors like defence, aerospace, automotive, renewable energy, and telecommunications. Countries increasingly seek access to reliable, secure, and resilient supplies of critical minerals, while global supply is uncertain due to market, technical, and commercial risks of exploration projects. This has made exploration geologists apply new technologies like artificial intelligence (AI) to increase the success rate of exploration projects. Recently, machine learning as a subset of AI has been successfully applied in different fields, such as spatial data analysis, to address different problems. This study proposes a machine learning-based framework for generating prospectivity maps of critical minerals focusing on the Gawler Craton in South Australia. This framework benefits from different novel machine learning methods for various purposes, including an improved generative adversarial network to overcome the class imbalance problem of the training dataset and the combination of positive and unlabelled learning and random forest as the main classifier for predicting mineralisation in the target area. We evaluated the efficiency of our proposed framework by creating prospectivity maps of mafic-ultramafic intrusion-hosted cobalt, chromium, and nickel mineralisation in the Gawler Craton. Various exploration datasets are used to generate input features, including publicly available geological, geophysical, and remote sensing datasets. We use known mineral occurrences as positive samples and randomly created a number of samples throughout the study area as unlabelled samples. Based on our results and different evaluation metrics, the model's performance is stable, and its accuracy is significantly higher than the model generated by a conventional approach using a standard random forest classifier. Our prospectivity maps show a strong spatial correlation between high probability values and known mineral occurrences and predicts several potential greenfield regions for future exploration.
... Harris et al., 2003;Agterberg, 2011), as well as machine learning techniques such as neural networks (e.g. Brown et al., 2000;Porwal et al., 2003;Tessema, 2017), random forests (Rodriguez-Galiano et al., 2014;Ford, 2020), and support vector machines (e.g. Zuo and Carranza, 2011). More recent developments in machine learning include the use of deep learning algorithms for assessing mineral potential (e.g. ...
The production of rare earth elements is critical for the transition to a low carbon economy. Carbonatites (>50% carbonate minerals) are one of the most significant sources of rare earth elements (REEs), both domestically within Australia, as well as globally. Given the strategic importance of critical minerals, including REEs, for the Australian national economy, a mineral potential assessment has been undertaken to evaluate the prospectivity for carbonatite-related REE (CREE) mineralisation in Australia. CREE deposits form as the result of lithospheric-to deposit-scale processes that are spatially and temporally coincident. Building on previous research into the formation of carbonatites and their related REE mineralisation, a mineral system model has been developed that incorporates four components: (1) source of metals, fluids, and ligands, (2) energy sources and fluid flow drivers, (3) fluid flow pathways and lithospheric architecture, and (4) ore deposition. This study demonstrates how national-scale datasets and a mineral systems-based approach can be used to map the mineral potential for CREE mineral systems in Australia. Using statistical analysis to guide the feature engineering and map weightings, a weighted index overlay method has been used to generate national-scale mineral potential maps that reduce the exploration search space for CREE mineral systems by up to ~90%. In addition to highlighting regions with known carbonatites and CREE mineralisation, the mineral potential assessment also indicates high potential in parts of Australia that have no previously identified carbonatites or CREE deposits.
... Various technical methods, such as drilling, geophysical exploration, geochemical exploration, and remote sensing, yield abundant geological data, serving as pivotal support for the application of machine learning and deep learning algorithms [17,18]. Artificial intelligence technology has found extensive use in quantitatively predicting mineral resources, encompassing neural networks [19], support vector machines [20][21][22], random forests [22][23][24][25][26][27], extreme learning machines [28,29], AHP [30,31], and more. Notably, deep learning and convolutional neural networks (CNN) have been widely employed in mineral prediction and classification. ...
Analyzing and fusing information layers of exploratory parameters is a critical step for enhancing the accuracy of identifying mineral potential zones during the reconnaissance stage of mineral exploration. The Qixia area in Shandong Province is characterized by intricate geological structures and abundant mineral resources. Numerous gold polymetallic deposits have been discovered in this region, highlighting the potential for discovering more such deposits in the ore concentration zone and its adjacent areas. In this study, we focus on the Qixia area and employ the box dimension method to analyze the fractal dimension of fault structures. We investigate the relationship between orebody occurrence and fault incidence within the mining region. Furthermore, we combine fractal analysis with Fry analysis to comprehensively predict the metallogenic potential in the area. This study reveals the fractal dimension values of fault structures, demonstrating that fault structures govern the distribution of ore bodies, with NE and NW fault structures being the primary ore-hosting features. Based on thorough analysis, we hypothesize that gold deposits in this area are generally distributed along the northeastern direction. By considering mineral distribution characteristics, this study identifies five potential metallogenic prospect areas within the study region. Capitalizing on advancements in information technology and big data, digital geology has gained prominence in prospecting and prediction. To this end, we construct a multi-information comprehensive prospecting model based on the structure-geochemical anomaly-mineralization alteration, employing the convolutional neural network (CNN) model for quantitative estimation of regional gold mineral resources. The findings validate the CNN model’s robust prediction performance in this area, leading to the determination of five prediction prospects. We observe a notable congruence between the two methods, offering significant insights for subsequent exploration endeavors in the region.
Graph-based models have been utilized for mineral prospectivity mapping (MPM), and they have demonstrated excellent performance owing to their adaptable graph structure, which is conducive to comprehensively considering the spatial anisotropy of mineralization compared with pixel- or image-based models. However, widely used graph-based models cannot fully consider the relationship between geological entities and mineralization. A heterogeneous graph is a type of graph structure containing rich heterogeneous information, allowing the consideration of various relationships and the assignment of suitable attributes to various types of nodes. Nodes in heterogeneous graphs can fully integrate heterogeneous information based on specific relations (i.e., edges). This study introduced a novel method for constructing heterogeneous graphs for MPM. The nodes in the graph consist of different types of geological entities, and the edges (relations) represent the links between the geological entities. The constructed heterogeneous graph cannot only effectively express the spatial anisotropy of mineralization but also consider the shape of geological entities and the relationships among geological entities, which is beneficial for modeling complex ore-forming geological processes. This heterogeneous graph was then trained using graph neural networks to obtain a mineral prospectivity map for southwestern Fujian Province, China. In addition, the proposed graph construction method demonstrated higher feasibility and accuracy in MPM compared to conventional graph construction method and convolutional neural networks.
Supervised machine learning algorithms are utilized to predict undiscovered mineral resources by analyzing the correlation between geological data and mineral deposits. The scarcity of mineralization and the uncertainty arising from the selection of training samples also the accuracy and generalization of such algorithms. This study employed the adaptive boosting (AdaBoost), gradient boosting decision tree (GBDT), and extreme gradient boosting (XgBoost) algorithms to map the prospectivity of tungsten polymetallic mineral resources in the Nanling metallogenic belt. Firstly, the under-sampling and synthetic minority oversampling technique (SMOTE) methods were used to generate training datasets. Secondly, 50 groups of training datasets were generated using under-sampling, and another 50 groups of training datasets were generated using the SMOTE method. These datasets were used to separately train different boosting algorithms in order to assess the uncertainty associated with the selection of negative samples and the generation of positive samples. Finally, the risk–return analysis was used to mitigate uncertainty, and an enhanced prediction–area (P–A) plot was proposed to evaluate the performance. The results indicate that AdaBoost is the least affected by the selection of negative samples, followed by XgBoost. The SMOTE not only enhances the performance of AdaBoost and XgBoost algorithms but it also reduces the uncertainty related to the selection of negative samples and the generation of positive samples. In addition, the enhanced P–A plot can simultaneously account for both prediction accuracy and uncertainty, making it a potential tool for model evaluation. According to the results, eight potential areas with high return and low risk have been identified as priority areas for exploration. This research not only introduces a new method for mineral prospectivity mapping and uncertainty evaluation but also provides guidance for mineral exploration in this region.
Global demand for critical raw materials, including phosphorus (P) and rare earth elements (REEs), is on the rise. The south part of Norway, with a particular focus on the Southern Oslo Rift region, is a promising reservoir of Fe-Ti-P-REE resources associated with magmatic systems. Confronting challenges in mineral exploration within these systems, notably the absence of alteration haloes and distal footprints, we have explored alternative methodologies. In this study, we combine machine learning with geological expertise, aiming to identify prospective areas for critical metal prospecting. Our workflow involves processing over 400 rock samples to create training datasets for mineralization and non-mineralization, employing an intuitive sampling strategy to overcome an imbalanced sample ratio. Additionally, we convert airborne magnetic, radiometric, and topographic maps into machine learning-friendly features, with a keen focus on incorporating domain knowledge into these data preparations. Within a binary classification framework, we evaluate two commonly used classifiers: a random forest (RF) and support vector machine (SVM). Our analysis shows that the RF model outperforms the SVM model. The RF model generates a predictive map, identifying approximately 0.3% of the study area as promising for mineralization. These findings align with legacy data and field visits, supporting the map’s potential to guide future surveys.
The purpose of mineral prospectivity mapping (MPM) is to discover unknown mineral deposits by means of fusing multisource prospecting information. In recent years, with rapid advancements in artificial intelligence, deep learning algorithms (DLAs) as a groundbreaking technique have exhibited outstanding capabilities in geoscience. However, conventional DLAs for MPM face certain challenges in feature extraction and the fusion of multimodal prospecting data. Moreover, opaque DLAs lead to an insufficient understanding of the predictive results by experts. In this study, a dual-branch convolutional neural network (DBCNN) and its post hoc interpretability were jointly constructed to map gold prospectivity in western Henan Province of China. In particular, channel and spatial attention modules were integrated into two branches to complement the respective advantages of multichannel and high spatial prospecting data for MPM. The Shapley additive explanations (SHAP) framework was then adopted to explain the predictive results by exploring the feature contributions. The comparative experiments illustrated that DBCNN can enhance feature representation and fusion abilities to improve the performance of MPM compared to conventional DLAs. The high-probability areas delineated by the DBCNN model exhibited close spatial relevance with known gold deposits, and the SHAP further confirmed the reliability of the predictive result obtained by the DBCNN model, thereby guiding future gold exploration in this study area.
1- Introduction
2- The study area of Shahre-Babak is a part of Urmia Dokhtar magmatic arc (UMDA). The extent of the study area is about 1977 km2, which is located in eastern part of Kerman province and approximately 170 kilometers far from Kerman city. The study area located on the 1:100,000 Shahre Babak, geology map which is high potential area for porphyry copper mineralization.
3- Material and methods
Text Predictive maps include nine layers of lithology information, lineaments, copper geochemical signature, multivariate signature anomaly resulting from factor analysis (factor 1), aeromagnetic data (reduction to pole), digital elevation model elevation, argillic alteration, phyllic alteration and iron oxide alteration. (Gossan zone). To extract the lithological layer of the study area, the Shahre Babak geology map, which was prepared by the Geological Survey of Iran, was used. The units extracted from the geological map of Shahre- Babak include sub-volcanic intrusive units, which are a suitable source for porphyry copper mineralization.
Lineaments are another effective parameter in porphyry copper mineralization. The effect of lineaments in porphyry copper mineralization has been investigated by various authors (Sillitoe, 1972, 1997; Skewes and Stern, 1994). The faults show a high tectonic activity and provide crushed zones suitable for porphyry copper mineralization. These places can be suitable location for the penetration of mineralized fluids and mineralization; Therefore, they can be considered as suitable keys for the recognition and exploration of mineral deposits. Therefore, studying the fractured zones and comparing the map of geochemical anomalies with the density map of lineaments can be useful in evaluating the anomalies.
The third layer used in finding high porphyry copper mineralization is the aeromagnetic data of Shahre-Babak, which were surveyed by the Atomic Energy Organization in 1977 with a line spacing of 500 meters and a height of 120 meters.
In regional exploration, stream sediment geochemistry is one of the steps to identify promising mineralization areas. One of the points in stream sediments geochemistry is evaluating the representativeness of a sample to predict the type of mineralization. In order to identify the promising areas of a specific type of mineralization, the best combination of trace elements should be identified and multivariate analysis should be used to achieve this goal.
The fourth layer of predictive maps is Aster satellite images. The mentioned images were downloaded from the United States Geological Survey (USGS) website. Argillic, phyllic and iron oxide alterations (Gossan zone) were extracted using band ratio methods.
One of the common methods in satellite image processing is the band ratio method. The application of the band ratio method is in the qualitative identification of mineralization zone related to hydrothermal alteration.
4- Adaptive Nero fuzzy method
The combination of fuzzy logic and neural network methods was first proposed by Jang (1993).
The combined method of fuzzy neural network, as its name suggests, uses the combination of two methods of neural network (data-oriented) and fuzzy logic (knowledge-oriented) in mineral potential modeling. This method can also be called knowledge-based neural network (Porwal et al., 2004). It uses a fuzzy inference system to form a matrix of eigenvectors at the input of the neural network. Therefore, the basic difference between the fuzzy neural network method and the neural network method is the way to form the matrix of eigenvectors.
5- Results and discussions
6- In order to train the model resulting from the adaptive nero fuzzy network in this research, two series of data are needed: The deposit points, which includes 38 points, are mineralized in the study area of Shahre-Babak area. These points entered the training model with index number one. 38 non-deposit points that were obtained using the point pattern analysis method, which were entered into the training model of the Adaptive Nero fuzzy network by index of number zero.
7- Conclusion
In this research, the adaptive Nero fuzzy method has been used in producing the porphyry potential model in the study area of Share- Babak. In this regard, nine exploratory criteria of subvolcanic units related to porphyry copper mineralization, faults, geochemical signature of copper element, geochemical signature of multivariate analysis (factor 1), aeromagnetic data, argillic alteration, phyllic alteration, Iron oxide alteration and DEM layer were used. Firstly, the mentioned layers were converted into a raster file and then these layers were transformed to same scaled using fuzzy transformations.
Next, information about 38 mineralization points and 38 non-mineralization points was extracted from the prepared data. Non-mineralization points were extracted using point pattern analysis method. The prepared training points were entered into MATLAB software with an index of one for mineralization points and zero index for non-mineralization points. After the training, the training model produced was implemented on the data of the study area and the final model was drawn out in the ArcGis software environment.
References
Sillitoe, H., 1997. Characteristic and controls of the largest porphyry copper–gold and epithermal gold deposits in the circum-pacific region. Australian Journal of Earth Sciences 44, 373–388.
https://doi.org/10.1080/08120099708728318
Sillitoe, R., 1972. A plate tectonic model for the origin of porphyry copper deposits. Economic Geology Journal 67, 184–197.
https://doi.org/10.2113/gsecongeo.67.2.184
Skewes,A., Stern, R., 1994. Tectonic trigger for the formation of late Miocene Cu-rich breccias pipe in the Andes of central Chile. Geology Journal 22, 551–554.
https://doi.org/10.1130/0091
Ninomiya,
Porwal, A., Carranza, E., Hale, M., 2004. A hybrid fuzzy weights-of evidence model for mineral potential mapping. Natural Resources Research Journal 15, 1–14.
https://doi.org/10.1007/s11053-006-9012-7
Taking the Pb–Zn-Fe deposit in Luziyuan area of Yunnan Province as the research object, the maximum entropy model was used to extract the prospecting information in the research area. The Pb–Zn-Fe deposits and relevant regional exploration data in this area were utilized to establish a model for mineral resource prediction, and the Pb–Zn-Fe prospecting prospect area in the research area was determined. The first and second components (PC1 and PC2) of the principal component scores (PC1 and PC2) of equidistant logarithmic transformation geochemical data were selected as model inputs. PC1 represented a major geochemical signature of the product of tectonic processes. PC2 was a secondary geochemical feature. Then, the importance of each evidence map was quantitatively ranked by principal component analysis. The response curves and marginal effect curves show that mineral exploration technology should be carried out in zoning according to the different metallogenic characteristics of Pb–Zn-Fe deposits. The experimental results show that the maximum entropy algorithm can be effectively applied to mineral resource exploration.
This contribution conceptualizes a blind source separation (BSS) model to recover sources of geochemical signals such that multi-depth ore-related enrichment patterns in complex metallogenic systems can be recognized. The proposed BSS framework consists of two consecutive modules. The first module is for the spectral decomposition of elemental mixtures to obtain different frequency-related components of signals induced by various geological sources. The second module serves to recover the sources of spectrally filtered geochemical signals according to the statistical assumptions made for the transmission of the latter from the former. In a real case experiment on a multiphase mineralization system, the proposed model was applied to the surface geochemical signals of ore-forming elements to gauge the relevance of source-related signals in depicting subsurface ore-related enrichment patterns. Multifractal filtering according to the generalized scale invariance characteristics of the power spectral density plane was adopted to derive elemental images enhanced in different spectral bands. Assuming linear instantaneous transmission, the FastICA technique was employed to encode spectrally filtered representations of elemental mixtures and recover source-related geochemical signals corresponding to different geo-processes. Support vector machines were used to train classifiers to establish statistical links between the surface geochemical signals and the shallow/deep ore-related enrichment patterns within the study area. The classification accuracies demonstrated that shallow/deep ore-related enrichment patterns can be recognized and distinguished more effectively using recovered source-related signals than using elemental mixtures or spectrally filtered representations. The results indicated that the proposed BSS model can provide efficient source-related geochemical signals to identify robust ore-related enrichment patterns with integrated grade and depth resolution to guide further metal exploration.
The demand for critical minerals is rapidly increasing worldwide, yet future global supply remains uncertain due to the difficulty in discovering new deposits using traditional methods. To increase the success rate of exploration projects for these vital resources, the use of artificial intelligence is continuously increasing for big and complex data analysis. This study proposes a new machine learning-based framework that tackles common problems associated with exploring critical mineral deposits, such as the shortage of known mineral occurrences, challenges in selecting negative samples in barren regions, and unbalanced training data. Our framework combines an improved generative adversarial network with positive and unlabelled learning to enhance efficiency. To test the performance of the framework, we create prospectivity maps of mafic–ultramafic intrusion-hosted mineralisation for cobalt, chromium, and nickel in the Gawler Craton, South Australia. The models are trained on a carefully selected set of independent features based on a conceptual model derived from open-access exploration data, resulting in high and stable performance. The prospectivity maps show a strong spatial correlation between high probabilities and known mineral occurrences and predict potential greenfield regions for future exploration. Our models demonstrate a significantly higher accuracy compared to a conventional approach using a standard random forest classifier and reveal that geophysical features play a crucial role in mapping prospective regions of critical minerals. Overall, our framework has the potential to significantly enhance critical mineral exploration by providing a more accurate and efficient approach to identifying prospective regions for future mining operations.
Mineral exploration, especially porphyry copper, plays a vital role in the sustainable green energy transition but is a challenging and risky operation. Studies have utilized various data like mineralogy, lithology, tectonics, geophysics, geochemistry, and remote sensing for porphyry copper exploration. However, previous hydrothermal alteration models were qualitative and lacked details, such as copper grade. This thesis introduces methodological advancements toward semi-quantitative and quantitative hydrothermal alteration models for predicting economic copper mineralization and assessing their accuracy and uncertainty sources.
The study region, the Kerman metallogenic belt in south-eastern Iran, is rich in different copper deposition types. The Kuh Panj porphyry copper deposit, an unmined area in this belt, served as the focal point for this research, using geological studies and various exploration datasets like rock samples, airborne measurements, and space-borne images.
Integration of different exploration evidence in the study region identified high-potential areas of copper mineralization, which were then compared with existing information to prioritize based on copper grade/tonnage. The accuracy of input data, particularly ASTER images for mineral mapping, was carefully assessed, showing that threshold selection holds greater importance than spatial accuracy. Uncertainty sources in the models were acknowledged, and mineral composition interpretation discrepancies between XRD and infrared reflectance spectroscopy were addressed due to detection limits and lack of threshold values. Modeling copper content required information on high-potential mineralization areas and considered accuracy and uncertainty sources. Using different linear regressions, copper content was modeled, enabling the discrimination of economic copper mineralization from gangue, while also revealing relationships between copper content, exploration evidence, and mineralogical information.
In conclusion, the thesis covered four main topics: (a) determining high-potential copper mineralization areas through exploration evidence integration, (b) assessing the accuracy of mineral maps derived from ASTER images, (c) addressing inconsistencies in mineral composition interpretation, and (d) modeling copper content and its relationships with minerals and exploration evidence. The study advocates for quantitative-predictive approaches to enhance porphyry copper exploration and facilitate the transition to sustainable green energy.
The support vector machine (SVM) is a group of theoretically superior machine learning algorithms. It was found competitive with the best available machine learning algorithms in classifying high-dimensional data sets. This paper gives an introduction to the theoretical development of the SVM and an experi- mental evaluation of its accuracy, stability and training speed in deriving land cover classié cations from satellite images. The SVM was compared to three other popular classié ers, including the maximum likelihood classié er (MLC), neural network classié ers (NNC) and decision tree classié ers (DTC). The impacts of kernel coné guration on the performance of the SVM and of the selection of training data and input variables on the four classié ers were also evaluated in this experiment.
A feedforward neural network with one hidden layer and five neurons was trained to recognize the distance to kuroko mineral deposits. Average amounts per hole of pyrite, sericite, and gypsum plus anhydrite as measured by X-rays in 69 drillholes were used to train the net. Drillholes near and between the Fukazawa, Furutobe, and Shakanai mines were used. The training data were selected carefully to represent well-explored areas where some confidence of the distance to ore was assured. A logarithmic transform was applied to remove the skewness of distance and each variable was scaled and centered by subtracting the median and dividing by the interquartile range. The learning algorithm of annealing plus conjugate gradients was used to minimize the mean squared error of the scaled distance to ore. The trained network then was applied to all of the 152 drillholes that had measured gypsum, sericite, and pyrite. A contour plot of the neural net predicted distance to ore shows fairly wide areas of 1 km or less to ore; each of the known deposit groups is within the 1 km contour. The high and low distances on the margins of the contoured distance plot are in part the result of boundary effects of the contouring algorithm. For example, the short distances to ore predicted west of the Shakanai (Hanaoka) deposits are in basement. However, the short distances to ore predicted northeast of Furotobe, just off the figure, coincide with the location of the Nurukawa kuroko deposit and the Omaki deposit, south of the Shakanai-Hanaoka deposits, seems to be on an extension of short distance to ore contour, but is beyond the 3 km limit from drillholes. Also of interest are some areas only a few kilometers from the Fukazawa and Shakanai groups of deposits that are estimated to be many kilometers from ore, apparently reflecting the network's recognition of the extreme local variability of the geology near some deposits.
The need to integrate large quantities of digital geoscience information to classify locations as mineral deposits or nondeposits has been met by the weights-of-evidence method in many situations. Widespread selection of this method may be more the result of its ease of use and interpretation rather than comparisons with alternative methods. A comparison of the weights-of-evidence method to probabilistic neural networks is performed here with data from Chisel Lake-Andeson Lake, Manitoba, Canada. Each method is designed to estimate the probability of belonging to learned classes where the estimated probabilities are used to classify the unknowns. Using these data, significantly lower classification error rates were observed for the neural network, not only when test and training data were the same (0.02 versus 23%), but also when validation data, not used in any training, were used to test the efficiency of classification (0.7 versus 17%). Despite these data containing too few deposits, these tests of this set of data demonstrate the neural network's ability at making unbiased probability estimates and lower error rates when measured by number of polygons or by the area of land misclassified. For both methods, independent validation tests are required to ensure that estimates are representative of real-world results. Results from the weights-of-evidence method demonstrate a strong bias where most errors are barren areas misclassified as deposits. The weights-of-evidence method is based on Bayes rule, which requires independent variables in order to make unbiased estimates. The chi-square test for independence indicates no significant correlations among the variables in the Chisel Lake–Andeson Lake data. However, the expected number of deposits test clearly demonstrates that these data violate the independence assumption. Other, independent simulations with three variables show that using variables with correlations of 1.0 can double the expected number of deposits as can correlations of –1.0. Studies done in the 1970s on methods that use Bayes rule show that moderate correlations among attributes seriously affect estimates and even small correlations lead to increases in misclassifications. Adverse effects have been observed with small to moderate correlations when only six to eight variables were used. Consistent evidence of upward biased probability estimates from multivariate methods founded on Bayes rule must be of considerable concern to institutions and governmental agencies where unbiased estimates are required. In addition to increasing the misclassification rate, biased probability estimates make classification into deposit and nondeposit classes an arbitrary subjective decision. The probabilistic neural network has no problem dealing with correlated variables—its performance depends strongly on having a thoroughly representative training set. Probabilistic neural networks or logistic regression should receive serious consideration where unbiased estimates are required. The weights-of-evidence method would serve to estimate thresholds between anomalies and background and for exploratory data analysis.
A Probabilistic Neural Network (PNN) was trained to classify mineralized and nonmineralized cells using eight geological, geochemical, and geophysical variables. When applied to a second (validation) set of well-explored cells that had been excluded from the training set, the trained PNN generalized well, giving true positive percentages of 86.7 and 93.8 for the mineralized and nonmineralized cells, respectively. All artifical neural networks and statistical models were analyzed and compared by the percentages of mineralized cells and barren cells that would be retained and rejected correctly respectively, for specified cutoff probabilities for mineralization. For example, a cutoff probability for mineralization of 0.5 applied to the PNN probabilities would have retained correctly 87.66% of the mineralized cells and correctly rejected 93.25% of the barren cells of the validation set. Nonparametric discriminant analysis, based upon the Epanechnikov Kernel performed better than logistic regression or parametric discriminant analysis. Moreover, it generalized well to the validation set of well-explored cells, particularly to those cells that were mineralized. However, PNN performed better overall than nonparametric discriminant analysis in that it achieved higher percentages of correct retention and correct rejection of mineralized and barren cells, respectively. Although the generalized regression neural network (GRNN) is not designed for a binary—presence or absence of mineralization— dependent variable, it also performed well in mapping favorability by an index valued on the interval [0, 1]. However, PNN outperformed GRNN in correctly retaining mineralized cells and rejecting barren cells of the validation set.
A Geographic Information System (GIS) is used to prepare and process digital geoscience data in a variety of ways for producing gold prospectivity maps of the Swayze greenstone belt, Ontario, Canada. Data used to produce these maps include geologic, geochemical, geophysical, and remotely sensed (Landsat). A number of modeling methods are used and are grouped into data-driven (weights of evidence, logistic regression) and knowledge-driven (index and Boolean overlay) methods. The weights of evidence (WofE) technique compares the spatial association of known gold prospects with various indicators (evidence maps) of gold mineralization, to derive a set of weights used to produce the final gold prospectivity map. Logistic regression derives statistical information from evidence maps over each known gold prospect and the coefficients derived from regression analysis are used to weight each evidence map. The gold prospectivity map produced from the index overlay process uses a weighting scheme that is derived from input by the geologist, whereas the Boolean method uses equally weighted binary evidence maps.
The resultant gold prospectivity maps are somewhat different in this study as the data comprising the evidence maps were processed purposely differently for each modeling method. Several areas of high gold potential, some of which are coincident with known gold prospects, are evident on the gold prospectivity maps produced using all modeling methods. The majority of these occur in mafic rocks within high strain zones, which is typical of many Archean greenstone belts.
The book documents and explains, in three parts, geochemical anomaly and mineral prospectivity mapping by using a geographic information system (GIS). Part I reviews and couples the concepts of (a) mapping geochemical anomalies and mineral prospectivity and (b) spatial data models, management and operations in a GIS. Part II demonstrates GIS-aided and GIS-based techniques for analysis of robust thresholds in mapping of geochemical anomalies. Part III explains GIS-aided and GIS-based techniques for spatial data analysis and geo-information sybthesis for conceptual and predictive modeling of mineral prospectivity. Because methods of geochemical anomaly mapping and mineral potential mapping are highly specialized yet diverse, the book explains only methods in which GIS plays an important role. The book avoids using language and functional organization of particular commercial GIS software, but explains, where necessary, GIS functionality and spatial data structures appropriate to problems in geochemical anomaly mapping and mineral potential mapping. Because GIS-based methods of spatial data analysis and spatial data integration are quantitative, which can be complicated to non-numerate readers, the book simplifies explanations of mathematical concepts and their applications so that the methods demonstrated would be useful to professional geoscientists, to mineral explorationists and to research students in fields that involve analysis and integration of maps or spatial datasets. The book provides adequate illustrations for more thorough explanation of the various concepts.
In the history of research of the learning problem one can extract four periods that can be characterized by four bright events: (i) Constructing the first learning machines, (ii) constructing the fundamentals of the theory, (iii) constructing neural networks, (iv) constructing the alternatives to neural networks.
A regional geochemical map interpolated from point data, usually sampled in surficial media such as stream sediments or lake sediments, may contain a large amount of information critical for mineral exploration and environmental studies. The geochemical map is, however, not ‘ready-to-use’ for such tasks as the determination of a local ‘anomaly’ or the characterization of a regional trend of one or more chemical elements as may be required for the purpose of mineral resource prediction. This becomes possible only after the map has been clearly divided into different components. Fractal filtering, a recently developed technique for decomposing a map or image into different components, helps to separate the anomaly from background or to extract other meaningful patterns from the geochemical map using both frequency and spatial information.
The fractal filters are formed by applying the fractal concentration-area model to the power spectrum of the processed geochemical field. They often constitute a group of irregularly shaped filters in the frequency domain that can separate the domain of wave numbers into distinct regions, each with a power spectrum following a similar power-law or fractal property. The corresponding patterns of the separate components are obtained after transformation back to the spatial domain.
The fractal filter can be applied to decompose the original geochemical field into a set of map components with distinct scaling ranges and anisotropy. The analysis of relationships among these decomposed maps can provide useful information for the interpretation and evaluation of anomalies or trends.
This paper briefly introduces the theory behind the fractal filtering technique. A case study of regional geochemical data of lake sediments from western Meguma Terrain, Southern Nova Scotia, Canada, is used to illustrate application of this technique to process the regional geochemical maps of the study area for the prediction of the turbidite-hosted gold deposits.
Bayesian weight-of-evidence and logistic regression models are implemented in a GIS environment for
regional-scale prospectivity modeling of greenstone belts in the Yilgarn Craton, Western Australia, for
magmatic nickel sulfide deposits. The input variables for the models consisted of derivative GIS layers that
were used as proxies for mappable exploration criteria for magmatic nickel sulfide deposits in the Yilgarn.
About 70% of the 165 known deposits of the craton were used to train the models; the remaining 30% was used to validate the models and, therefore, had to be treated as if they had not been discovered. The weights-ofevidence and logistic regression models, respectively, classify 71.4% and 81.6% validation deposits in prospective zones that occupy about 9% of the total area occupied by the greenstone belts in the craton. The
superior performance of the logistic regression model is attributed to its capability to accommodate conditional
dependencies amongst the input predictor maps, and provide less biased estimates of prospectivity.
Setting of the learning problem consistency of learning processes bounds on the rate of convergence of learning processes controlling the generalization ability of learning processes constructing learning algorithms what is important in learning theory?.
Accurate mapping of potentially mineralized zones is best undertaken on the basis of multi-source exploration data-sets. Mineral potential mapping is more difficult when only geological data are available. A 'wildcat' method of predictive mapping is presented for use in this situation. Scores of proximity classes of indicative geological features are employed in principal-component analysis to extract a favourability function that can be interpreted as a representation of mineral potential. The method was tested in the Baguio district of the Philippines. Knowledge of the spatial associations between the gold deposits and the indicative geological features in the test district was not applied in the predictive mapping. The maps delineated some 60% of the known gold deposits correctly.
Wildcat modelling of mineral prospectivity has been proposed for greenfields geologically-permissive terranes where mineral targets have not yet been discovered but a geological map is available as a source of spatial data of predictors of mineral prospectivity. This paper (i) revisits the initial way of assigning wildcat scores (Sc) to predictors of mineral prospectivity and (ii) proposes an improvement by transforming Sc into improved wildcat scores (ISc) by using a logistic function. This was shown in wildcat modelling of prospectivity for low-sulphidation epithermal-Au (LSEG) deposits in Aroroy district (Philippines). Based on knowledge of characteristics of and controls on LSEG mineralization in the Philippines, the spatial predictors of LSEG prospectivity used in the study are proximity to porphyry plutonic stocks, faults/fractures and fault/fracture intersections. The Sc and ISc of spatial predictors are input separately to principal components analysis to extract a favourability function that can be interpreted as a wildcat model of LSEG prospectivity. The predictive capacity of the wildcat model of LSEG prospectivity based on the ISc of geological predictors is roughly 70% higher than that of the wildcat model of LSEG prospectivity based on the Sc of geological predictors. A slight increase of predictive capacity of wildcat modelling of LSEG prospectivity is also achieved when the ISc of geological predictors are integrated with the ISc of geochemical anomalies, but not with the Sc of geochemical anomalies. The proposed improvement is significant because if the study district were a greenfields exploration area, then a wildcat model of LSEG prospectivity based on the old wildcat methodology would have caused several LSEG targets to be missed.
A case application of a hybrid method combining multilevel fuzzy comprehensive evaluation with asymmetric fuzzy relation analysis is demonstrated to map porphyry-copper prospectivity in the Gangdese district, Tibet, western China. Two-level binary geoscience variables including favorable rocks, intrusive rocks, faults and geochemical anomalies, are identified based on porphyry-copper mineral deposit model and ranked using asymmetric fuzzy relation analysis. The prospective target areas are identified by fuzzy membership values greater than 0.82 and corresponding uncertainty less than 0.36. The target areas occupy only 7.0% of the study area but contain 52.4% of the total number of known deposits. The prospectivity map illustrates that the targets are not only directly associated with the known deposits in the central part of the study area, but also with E–W and N–E oriented faults in the northern part of the study area. The targets can be used to guide further exploration for undiscovered porphyry copper in Gangdese district.
A case application of data-driven estimation of evidential belief functions (EBFs) is demonstrated to prospectivity mapping in Lundazi district (eastern Zambia). Spatial data used to represent recognition criteria of prospectivity for aquamarine-bearing pegmatites include mapped granites, mapped faults/fractures, mapped shear zones, and radioelement concentration ratios derived from gridded airborne radiometric data. Data-driven estimates EBFs take into account not only (a) spatial association between an evidential map layer and target deposits but also (b) spatial relationships between classes of evidences in an evidential map layer. Data-driven estimates of EBFs can indicate which spatial data provide positive or negative evidence of prospectivity. Data-driven estimates of EBFs of only spatial data providing positive evidence of prospectivity were integrated according to Dempster’s rule of combination. Map of integrated degrees of belief was used to delineate zones of relative degress of prospectivity for aquamarine-bearing pegmatites. The predictive map has at least 85% prediction rate and at least 79% success rate of delineating training and validation deposits, respectively. The results illustrate usefulness of data-driven estimation of EBFs in GIS-based predictive mapping of mineral prospectivity. The results also show usefulness of EBFs in managing uncertainties associated with evidential maps.
Singular physical or chemical processes may result in anomalous amounts of energy release or mass accumulation that, generally,
are confined to narrow intervals in space or time. Singularity is a property of different types of non-linear natural processes
including cloud formation, rainfall, hurricanes, flooding, landslides, earthquakes, wildfires, and mineralization. The end
products of these non-linear processes can be modeled as fractals or multifractals. Hydrothermal processes in the Earth’s
crust can result in ore deposits characterized by high concentrations of metals with fractal or multifractal properties. Here
we show that the non-linear properties of the end products of singular mineralization processes can be applied for prediction
of undiscovered mineral deposits and for quantitative mineral resource assessment, whether for mineral exploration or for
regional, national and global planning for mineral resource utilization. In addition to the general theory and framework for
the non-linear mineral resources assessment, this paper focuses on several power-law models proposed for characterizing non-linear
properties of mineralization and for geoinformation extraction and integration. The theories, methods, and computer system
discussed in this paper were validated using a case study dealing with hydrothermal Au mineral potential in southern Nova
Scotia, Canada.
This paper describes a GIS-based application of a radial basis functional link net (RBFLN) to map the potential of SEDEX-type base metal deposits in a study area in the Aravalli metallogenic province (western India). Available public domain geodata of the study area were processed to generate evidential maps, which subsequently were encoded and combined to derive a set of input feature vectors. A subset of feature vectors with known targets (i.e., either known mineralized or known barren locations) was extracted and divided into (a) a training data set and (b) a validation data set. A series of RBFLNs were trained to determine the network architecture and estimate parameters that mapped the maximum number of validation vectors correctly to their respective targets. The trained RBFLN that gave the best performance for the validation data set was used for processing all feature vectors. The output for each feature vector is a predictive value between 1 and 0, indicating the extent to which a feature vector belongs to either the mineralized or the barren class. These values were mapped to generate a predictive classification map, which was reclassified into a favorability map showing zones with high, moderate and low favorability for SEDEX-type base metal deposits in the study area. The method demarcates successfully high favorability zones, which occupy 6% of the study area and contain 94% of the known base metal deposits.
Abstract
This study compares the performance of favorability mappings by weights of evidence (WOE), probabilistic neural networks (PNN), logistic regression (LR), and discriminant analysis (DA). Comparisons are made by an objective measure of performance that is based on statistical decision theory. The study further emphasizes out-of-sample inference, and quantifies the extent to which outcome is influenced by optimum variable discretization with classification and regression trees (CARTS).
Favorability mapping methodologies are evaluated systematically across three case studies with contrasting scale and geologic information:
Case Study Carlin Alamos Nevada
sediment-hosted intrusion-related intrusion-related
gold copper copper
Scale deposit district regional
Cell size small (0.01 km2) medium (1 km2) large (7 km2)
Information level high moderate low
Geovariables complex simple simple
Variable interdependency moderate low high
Asymmetry in frequency of barren and mineralized cells modest considerable severe
Estimated favorabilities for all cells then are represented by computed percent correct classification, and expected loss of optimum decision.
The deposit-scale Carlin study reveals that the performances of the various methods from lowest to highest expected decision loss are: PNN, nonparametric DA, binary PNN (WOE variables), LR, and WOE. Moreover, the study indicates that approximately 40% of the increase in expected decision loss using WOE instead of PNN is the result of information loss from variable discretization. The remaining increases in losses using WOE are the result of its lesser inferential power than PNN.
The district-scale Alamos study shows that the lowest expected decision loss is not by PNN, but by canonical DA. CARTS discretization improves greatly the performance of WOE. However, PNN and DA perform better than WOE.
Unlike findings from the Alamos and Carlin studies, results from the regional-scale Nevada study indicate that decision losses by LR and DA are lower than those by WOE or PNN.
This paper proposes a new approach of weights of evidence method based on fuzzy sets and fuzzy probabilities for mineral potential mapping. It can be considered as a generalization of the ordinary weights of evidence method, which is based on binary or ternary patterns of evidence and has been used in conjunction with geographic information systems for mineral potential mapping during the past few years. In the newly proposed method, instead of separating evidence into binary or ternary form, fuzzy sets containing more subjective genetic elements are created; fuzzy probabilities are defined to construct a model for calculating the posterior probability of a unit area containing mineral deposits on the basis of the fuzzy evidence for the unit area. The method can be treated as a hybrid method, which allows objective or subjective definition of a fuzzy membership function of evidence augmented by objective definition of fuzzy or conditional probabilities. Posterior probabilities calculated by this method would depend on existing data in a totally data-driven approach method, but depend partly on expert's knowledge when the hybrid method is used. A case study for demonstration purposes consists of application of the method to gold deposits in Meguma Terrane, Nova Scotia, Canada.
A GIS-based hybrid neuro-fuzzy approach to mineral potential mapping implements a Takagi–Sugeno type fuzzy inference system in a four-layered feed-forward adaptive neural network. In this approach, each unique combination of predictor patterns is considered a feature vector whose components are derived by knowledge-based ordinal encoding of the constituent predictor patterns. A subset of feature vectors with a known output target vector (i.e., unique conditions known to be associated with either a mineralized or a barren location), extracted from a set of all feature vectors, is used for the training of an adaptive neuro-fuzzy inference system. Training involves iterative adjustment of parameters of the adaptive neuro-fuzzy inference system using a hybrid learning procedure for mapping each training vector to its output target vector with minimum sum of squared error. The trained adaptive neuro-fuzzy inference system is used to process all feature vectors. The output for each feature vector is a value that indicates the extent to which a feature vector belongs to the mineralized class or the barren class. These values are used to generate a favorability map. The procedure is applied to regional-scale base metal potential mapping in a study area located in the Aravalli metallogenic province (western India). The adaptive neuro-fuzzy inference system demarcates high favorability zones occupying 9.75% of the study area and identifies 96% of the known base metal deposits. This result is significant both in terms of reduction in search area and the percentage of deposits identified.
This paper describes a hybrid fuzzy weights-of-evidence (WofE) model for mineral potential mapping that generates fuzzy predictor
patterns based on (a) knowledge-based fuzzy membership values and (b) data-based conditional probabilities. The fuzzy membership
values are calculated using a knowledge-driven logistic membership function, which provides a framework for treating systemic
uncertainty and also facilitates the use of multiclass predictor maps in the modeling procedure. The fuzzy predictor patterns
are combined using Bayes’ rule in a log-linear form (under an assumption of conditional independence) to update the prior
probability of target deposit-type occurrence in every unique combination of predictor patterns. The hybrid fuzzy WofE model
is applied to a regional-scale mapping of base-metal deposit potential in the south-central part of the Aravalli metallogenic
province (western India). The output map of fuzzy posterior probabilities of base-metal deposit occurrence is classified subsequently
to delineate zones with high-favorability, moderate favorability, and low-favorability for occurrence of base-metal deposits.
An analysis of the favorability map indicates (a) significant improvement of probability of base-metal deposit occurrence
in the high-favorability and moderate-favorability zones and (b) significant deterioration of probability of base-metal deposit
occurrence in the low-favorability zones. The results demonstrate usefulness of the hybrid fuzzy WofE model in representation
and in integration of evidential features to map relative potential for mineral deposit occurrence.
In the central California coastal forests, a newly discovered virulent pathogen (Phytophthora ramorum) has killed hundreds of thousands of native oak trees. Predicting the potential distribution of the disease in California remains an urgent demand of regulators and scientists. Most methods used to map potential ranges of species (e.g. multivariate or logistic regression) require both presence and absence data, the latter of which are not always feasibly collected, and thus the methods often require the generation of ‘pseudo’ absence data. Other methods (e.g. BIOCLIM and DOMAIN) seek to model the presence-only data directly. In this study, we present alternative methods to conventional approaches to modeling by developing support vector machines (SVMs), which are the new generation of machine learning algorithms used to find optimal separability between classes within datasets, to predict the potential distribution of Sudden Oak Death in California. We compared the performances of two types of SVMs models: two-class SVMs with ‘pseudo’ absence data and one-class SVMs. Both models performed well. The one-class SVMs have a slightly better true-positive rate (0.9272 ± 0.0460 S.D.) than the two-class SVMs (0.9105 ± 0.0712 S.D.). However, the area predicted to be at risk for the disease using the one-class SVMs (18,441 km2) is much larger than that of the two-class SVMs (13,828 km2). Both models show that the majority of disease risk will occur in coastal areas. Compared with the results of two-class SVMs, the one-class SVMs predict a potential risk in the foothills of the Sierra Nevada mountain ranges; much greater risks are also found in Los Angles and Humboldt Counties. We believe the support vector machines when coupled with geographic information system (GIS) will be a useful method to deal with presence-only data in ecological analysis over a range of scales.
This paper shows the use of a data domain description method, inspired by the support vector machine by Vapnik, called the support vector domain description (SVDD). This data description can be used for novelty or outlier detection. A spherically shaped decision boundary around a set of objects is constructed by a set of support vectors describing the sphere boundary. It has the possibility of transforming the data to new feature spaces without much extra computational cost. By using the transformed data, this SVDD can obtain more flexible and more accurate data descriptions. The error of the first kind, the fraction of the training objects which will be rejected, can be estimated immediately from the description without the use of an independent test set, which makes this method data efficient. The support vector domain description is compared with other outlier detection methods on real data.
Information about the Earth's surface is required in many wide-scale applications. Land cover/use classification using remotely sensed images is one of the most common applications in remote sensing, and many algorithms have been developed and applied for this purpose in the literature. Support vector machines (SVMs) are a group of supervised classification algorithms that have been recently used in the remote sensing field. The classification accuracy produced by SVMs may show variation depending on the choice of the kernel function and its parameters. In this study, SVMs were used for land cover classification of Gebze district of Turkey using Landsat ETM+ and Terra ASTER images. Polynomial and radial basis kernel functions with their estimated optimum parameters were applied for the classification of the data sets and the results were analyzed thoroughly. Results showed that SVMs, especially with the use of radial basis function kernel, outperform the maximum likelihood classifier in terms of overall and individual class accuracies. Some important findings were also obtained concerning the changes in land use/cover in the study area. This study verifies the effectiveness and robustness of SVMs in the classification of remotely sensed images.
This paper explores the use of Support Vector Machines (SVMs) for learning text classifiers from examples. It analyzes the particular properties of learning with text data and identifies why SVMs are appropriate for this task. Empirical results support the theoretical findings. SVMs achieve substantial improvements over the currently best performing methods and behave robustly over a variety of different learning tasks. Furthermore they are fully automatic, eliminating the need for manual parameter tuning.
A data-driven application of the theory of evidential belief to map mineral potential is demonstrated with a redefinition of procedures to estimate evidential belief functions. The redefined estimates of evidential belief functions take into account not only the spatial relationship of an evidence with the target mineral deposit but also consider the relationships among the subsets of spatial evidences within a set of evidential data layer. Proximity of geological features to mineral deposits is translated into spatial evidence and evidential belief functions are estimated for the proposition that mineral deposits exist in a test area. The integrated maps of degrees of belief for the proposition that mineral deposits exist in a test area is classified into a binary mineral potential map. For the Baguio district (Philippines), the binary gold potential map delineates (a) about 74% of the training data (i.e., locations of large-scale gold deposits) and (b) about 64% of the validation data (i.e., locations of small-scale gold deposits). The results demonstrate the usefulness of a geologically constrained mineral potential mapping using data-driven evidential belief functions to guide further surficial exploration work in the search for yet undiscovered gold deposits in the Baguio district. The results also indicate the usefulness of evidential belief functions for mapping uncertainties in the geologically constrained integrated predictive model of gold potential.
Data-driven prospectivity mapping can be undermined by dissimilarity in multivariate spatial data signatures of deposit-type locations. Most cases of data-driven prospectivity mapping, however, make use of training sets of randomly selected deposit-type locations with the implicit assumption that they are coherent (i.e., with similar multivariate spatial data signatures). This study shows that the quality of data-driven prospectivity mapping can be improved by using a training set of coherent deposit-type locations. Analysis and selection of coherent deposit-type locations was performed via logistic regression, by using multiple sets of deposit occurrence favourability scores of univariate geoscience spatial data as independent variables and binary deposit occurrence scores as dependent variable. The set of coherent deposit-type locations and three sets of randomly selected deposit-type locations were each used in data-driven prospectivity mapping via application of evidential belief functions. The prospectivity map based on the training set of coherent deposit-type locations resulted in lower uncertainty, better goodness-of-fit to the training set, and better predictive capacity against a cross-validation set of economic deposits of the type sought. This study shows that explicit selection of training set of coherent deposit-type locations should be applied in data-driven prospectivity mapping.
Reimpresión en 2002 Incluye bibliografía e índice
Kernel methods, a new generation of learning algorithms, utilize techniques from optimization, statistics, and functional analysis to achieve maximal generality, flexibility, and performance. These algorithms are different from earlier techniques used in machine learning in many respects: For example, they are explicitly based on a theoretical model of learning rather than on loose analogies with natural learning systems or other heuristics. They come with theoretical guarantees about their performance and have a modular design that makes it possible to separately implement and analyze their components. They are not affected by the problem of local minima because their training amounts to convex optimization. In the last decade, a sizable community of theoreticians and practitioners has formed around these methods, and a number of practical applications have been realized. Although the research is not concluded, already now kernel methods are considered the state of the art in several machine learning tasks. Their ease of use, theoretical appeal, and remarkable performance have made them the system of choice for many learning problems. Successful applications range from text categorization to handwriting recognition to classification of gene-expression data.
ArcWofE: ArcView extension for weights of evidence mapping
- L D Kemp
- G F Bonham-Carter
- G L Raines
Kemp, L.D., Bonham-Carter, G.F., Raines, G.L., 1999. ArcWofE: ArcView extension for
weights of evidence mapping: /http://gis.nrcan.gc.ca/software/arcview/
wofeS.
Information Synthesis for Mineral Exploration
- G Pan
- D P Harris
Pan, G., Harris, D.P., 2000. Information Synthesis for Mineral Exploration. Oxford
Univ. Press, New York, 461 pp.
A Tutorial Guide to using MI-SDM v2.50 based on USGS Open-File Report 01-221 by Geographic Information Systems for Geoscientists: Modelling with GIS Weights of evidence modelling: a new approach to mapping mineral potential
- Avantra
Avantra Geosystems, 2006. A Tutorial Guide to using MI-SDM v2.50 based on USGS Open-File Report 01-221 by Gary L. Raines. Bonham-Carter, G.F., 1994. Geographic Information Systems for Geoscientists: Modelling with GIS. Pergamon, Ontario, 398 pp. Bonham-Carter, G.F., Agterberg, F.P., Wright, D.F., 1989. Weights of evidence modelling: a new approach to mapping mineral potential. In: Agterberg, F.P., Bonham-Carter, G.F. (Eds.), Statistical Applications in the Earth Sciences. Geological Survey of Canada, Paper 89-9, pp. 171–183.
Metallogenic map of Nova Scotia, version 1, scale 1:500 000. Department of Mines and Energy, Nova Scotia GeoData Analysis System (GeoDAS) for mineral Exploration: User's Guide and Exercise Manual
- A K Chatterjee
- Canada
- Q Cheng
Chatterjee, A.K., 1983. Metallogenic map of Nova Scotia, version 1, scale 1:500 000. Department of Mines and Energy, Nova Scotia, Canada. Cheng, Q., 2000. GeoData Analysis System (GeoDAS) for mineral Exploration: User's Guide and Exercise Manual. Material for the training workshop on GeoDAS held at York University, Toronto, Canada, 1, 3, 204, /http://www.gisworld.org/ geodasS.
ArcMap 9.3 geoprocessing tools for spatial modeling using weights of evidences, logistic regression
- D I Sawatzky
- G L Raines
- G F Bonham-Carter
- Looney
Sawatzky, D.I., Raines, G.L., Bonham-Carter, G.F., Looney, 2009. Spatial Data Modeller
(SDM): ArcMap 9.3 geoprocessing tools for spatial modeling using weights of
evidences, logistic regression, fuzzy logic and neural networks. /http://arc
scripts.esri.com/details.asp/dbid=15341S.