No full-text available
To read the full-text of this research,
you can request a copy directly from the authors.
Support vector regression
... Similar to the classification problem, the Lagrange function of the main problem is given by [46]: ...
... Additionally, contrary to the classification case, there is no assumption as = 1, because in regression, {yi} determines the size of the problem. Now, taking the derivatives of the Lagrange function with respect to w, b, £ * £ , we have [46]: ...
... They are obtained with the equation = ∑ (∝ −∝ * ) for the coefficients matrix, which is made up of input data xi and is independent of the state variable yi. To calculate b, the following equations can be used [46]: Figure 4 illustrates the process of proposed model from start to stop. ...
The global population growth has led to a considerable rise in demand for wheat. Today, the amount of energy consumption in agriculture has also increased due to the need for sufficient food for the growing population. Thus, agricultural policymakers in most countries rely on prediction models to influence food security policies. This research aims to predict and reduce the amount of energy consumption in wheat production. Data were collected from the farms of Estahban city in Fars province of Iran by the Jihad Agricultural Department's experts for 20 years from 1994 to 2013. In this study, a novel prediction method based on consumed energy in the production period is proposed. The model is developed based on artificial intelligence to forecast the output energy in wheat production and uses extreme learning machine (ELM) and support vector regression (SVR). In the experimental stage, the value of elevation metrics for the EVM and ELM was reported to be equal to 0.000000409 and 0.9531, respectively. Total input energy (consumed) is found to be 1,460,503.1 Mega Joules (MJ), and output energy (produced wheat) is 1,401,011.945 MJ for the Estahban. The result indicates the superiority of the ELM model to enhance the decisions of the agricultural policymakers.
... A stream's "low flow" refers to the amount of water flowing in a stream during prolonged periods of little to no rainfall during an average non-drought year. The low-flow regime for a particular stream is controlled by the physical characteristics of its basin and the local climate (Smakhtin, 2001). The 7Q10 statistic describes a basin's expected low-flow and provides a way to compare directly the low-flow regimes of different basins. ...
... In addition to variable importance, we explored the effect of each predictor in more detail using partial dependence plots (Fig. 7, right panel). In a review of low-flow hydrology, Smakhtin (2001) lists several factors that influence the low-flow regime of a basin: the distribution and infiltration characteristics of soils, the hydraulic characteristics and extent of the aquifers, the rate, frequency and amount of recharge, the evapotranspiration rates from the basin, distribution of vegetation types, topography and climate. Five of the seven factors mentioned by Smakhtin (2001), including soils, aquifer characteristics, recharge, vegetation type, and climate are reflected in the most important predictor variables identified here for the machine-learning models (Fig. 7). ...
... In a review of low-flow hydrology, Smakhtin (2001) lists several factors that influence the low-flow regime of a basin: the distribution and infiltration characteristics of soils, the hydraulic characteristics and extent of the aquifers, the rate, frequency and amount of recharge, the evapotranspiration rates from the basin, distribution of vegetation types, topography and climate. Five of the seven factors mentioned by Smakhtin (2001), including soils, aquifer characteristics, recharge, vegetation type, and climate are reflected in the most important predictor variables identified here for the machine-learning models (Fig. 7). ...
We compare the ability of eight machine-learning models (elastic net, gradient boosting, kernel-k-nearest neighbors, two variants of support vector machines, M5-cubist, random forest, and a meta-learning ensemble M5-cubist model) and four baseline models (ordinary kriging, a unit area discharge model, and two variants of censored regression) to generate estimates of the annual minimum 7-day mean streamflow with an annual exceedance probability of 90% (7Q10) at 224 unregulated sites in South Carolina, Georgia, and Alabama, USA. The machine-learning models produced substantially lower cross validation errors compared to the baseline models. The meta-learning M5-cubist model had the lowest root-mean-squared-error of 26.72 cubic feet per second. Partial dependence plots show that 7Q10s are likely moderated by late summer and early fall precipitation and the infiltration capacity of basin soils.
... This means that as long as the error of the predicted value does not exceed ε, it is considered that the error is zero. If it exceeds ε it is calculated as 'error-ε', giving a different solution to the problem (Parrella 2007). Thus, the predicted values of Online-SVR can be transformed into the following quadratic optimization problem (Parrella 2007): ...
... If it exceeds ε it is calculated as 'error-ε', giving a different solution to the problem (Parrella 2007). Thus, the predicted values of Online-SVR can be transformed into the following quadratic optimization problem (Parrella 2007): ...
... To use the Online-SVR model, a set of initial samples must be obtained and input into the model for training. When additional sample data need to be added to the model, the following inputs to the program are required (Parrella 2007): (1) the trained Online-SVR model; (2) (α i − α * i ) and b of the trained Online-SVR; (3) Online-SVR parameters, kernel type and kernel parameters; and (4) new samples. The Online-SVR then updates (α i − α * i ) and b while adhering to the constraints that were set initially. ...
The structural crashworthiness design of vehicles has become an important research direction to ensure the safety of the occupants. To effectively improve the structural safety of a vehicle in a frontal crash, a system methodology is presented in this study. The surrogate model of Online support vector regression (Online-SVR) is adopted to approximate crashworthiness criteria and different kernel functions are selected to enhance the accuracy of the model. The Online-SVR model is demonstrated to have the advantages of solving highly nonlinear problems and saving training costs, and can effectively be applied for vehicle structural crashworthiness design. By combining the non-dominated sorting genetic algorithm II and Monte Carlo simulation, both deterministic optimization and reliability-based design optimization (RBDO) are conducted. The optimization solutions are further validated by finite element analysis, which shows the effectiveness of the RBDO solution in the structural crashworthiness design process. The results demonstrate the advantages of using RBDO, resulting in not only increased energy absorption and decreased structural weight from a baseline design, but also a significant improvement in the reliability of the design.
... An E. coli bacterium in contact with a film of antibody functionalised graphene is illustrated in Fig. 1. In this configuration, to check the system response and to find the kinetics of bacteria binding, the sensor was kept in chambers with 10 5 cfu/ml of E. coli [4,38]. As the higher number of E. coli is caught by the graphene film antibodies, the conductance of the channel increases. ...
... In our research, we use the ε-SVR, which gives an approximate function f(x) that limits the deviations from the target y i in the training data set (i.e. {(x 1 , y 1 ),…,(x I , y L )} ⊆ (X × Y ) l ) not to exceed a maximum value ε and makes it as flat as possible [3,38]. In other words, errors are neglected as long as they are smaller than a pre-specified value, ε. ...
Graphene is an allotrope of carbon with two-dimensional (2D) monolayer honeycombs. A larger detection area and higher sensitivity can be provided by graphene-based nanosenor because of its 2D structure. In addition, owing to its special characteristics, including electrical, optical and physical properties, graphene is known as a more suitable candidate compared to other materials used in the sensor application. A novel model employing a field-effect transistor structure using graphene is proposed and the current-voltage (I-V) characteristics of graphene are employed to model the sensing mechanism. This biosensor can detect Escherichia coli (E. coli) bacteria, providing high levels of sensitivity. It is observed that the graphene device experiences a drastic increase in conductance when exposed to E. coli bacteria at 0-105 cfu/ml concentration. The simple, fast response and high sensitivity of this nanoelectronic biosensor make it a suitable device in screening and functional studies of antibacterial drugs and an ideal high-throughput platform which can detect any pathogenic bacteria. Artificial neural network and support vector regression algorithms have also been used to provide other models for the I-V characteristic. A satisfactory agreement has been presented by comparison between the proposed models with the experimental data.
... Further, these models are strictly parametric and must satisfy several restrictive assumptions on data distribution. In this study, we propose the application of support vector machines (SVM) (Vapnik 1995(Vapnik , 1998Cherkassky and Mullier 2007) to landslide susceptibility mapping. SVM are based on a non-linear transformation of the covariates in a high-dimensional space, where different classes are linearly separable. ...
... In other words, SVM are data-dependent models, which means that the model capacity is tuned to match data complexity. This paradigm, also termed: structural risk minimization (Vapnik 1995;Cherkassky and Mullier 2007), is the basis of the SVM learning algorithm. In this study, our aim is to discriminate between susceptible (1) and not susceptible (−1) pixels. ...
The aim of this study is the application of support vector machines (SVM)
to landslide susceptibility mapping. SVM are a set of machine learning methods in
which model capacity matches data complexity. The research is based on a conceptual
framework targeted to apply and test all the procedural steps for landslide susceptibility
modeling from model selection, to investigation of predictive variables,
from empirical cross-validation of results, to analysis of predicted patterns. SVM
were successfully applied and the final susceptibility map was interpreted via success
and prediction rate curves and receiver operating characteristic (ROC) curves, to
support the modeling results and assess the robustness of the model. SVM appeared
to be very specific learners, able to discriminate between the informative input and
random noise. About 78% of occurrences was identified within the 20% of the most
susceptible study area for the cross-validation set. Then the final susceptibility map
was compared with other maps, addressed by different statistical approaches, commonly
used in susceptibility mapping, such as logistic regression, linear discriminant
analysis, and naive Bayes classifier. The SVM procedure was found feasible and
able to outperform other techniques in terms of accuracy and generalization capacity.
The over-performance of SVM against the other techniques was around 18% for the
cross-validation set, considering the 20% of the most susceptible area. Moreover, by
analyzing receiver operating characteristic (ROC) curves, SVM appeared to be less
prone to false positives than the other models. The study was applied in the Staffora
river basin (Lombardy, Northern Italy), an area of about 275 km2 characterized by
avery high density of landslides, mainly superficial slope failures triggered by intense
rainfall events.
... SVM became a popular approach when excellent results where achieved in classification and function approximation. SVM perfor-mances are shown to outperform those of ANN and other statistical methods on several benchmark problems (Meyer et al., 2003;van der Walt and Barnard, 2006;Cherkassky and Mullier, 2007a). ...
... Another consideration about SVR tuning is to choose a high value for C and keep it fixed while tuning the value of ε (Cherkassky and Mullier, 2007a). In this way the insensitivity tube wideness is adjusted until the optimal value is reached. ...
Digital soil mapping in mountain areas faces two major limitations: the small number of available observations and the non-linearity of the relations between environmental variables and soil properties.A possible approach to deal with these limitations involves the use of non-parametric models to interpolate soil properties of interest. Among the different approaches currently available, Support Vector Regression (SVR) seems to have several advantages over other techniques. SVR is a set of techniques in which model complexity is limited by the learning algorithm itself, which prevents overfitting. Moreover, the non-linear approximation of SVR is based on a kernel transformation of the data, which avoids the use of complex functions and is computationally feasible; while the resulting projection in feature space is especially suited for sparse datasets.A brief introduction to this methodology, a comparison with other popular methodologies and a framework for the application of this approach to a study site in the Italian Alps is discussed.
... This ability to solve nonlinear regression problems was only possible with the introduction of Vapnik's ϵ-insensitive 1 loss function, an ability highlighted by its higher performance than other conventional regression techniques, as we can see from the articles by Mukherjee et al. (1997) and Müller et al. (1997). The optimization problem for the nonlinear SVR case is formulated as described in Smola and Schölkopf (2004), Welling (2004): ...
This paper aims to highlight in a relevant way the interest of hybrid models (coupling of ARIMA processes and machine learning models) for economic or financial agents. These models are likely to allow better consideration of certain stylized facts (not necessarily taken into account by often-used models such as ARIMA-GARCH), observed in the analysis of financial time series. Most hybrid models assume that the random perturbation of the chosen ARIMA process follows a normal distribution. Thus, the literature on hybrid models remains on this Gaussian framework and it would be necessary to consider a non-Gaussian and much more realistic framework. We consider hybrid models with the assumption that the random disturbance process follows a Student’s distribution (denoted by ARIMA T), allowing us to take into account the leptokurticity often observed when analyzing financial time series returns. Under certain assumptions, the empirical results show the power of the hybrid models considered.
... When data is not linearly separable, kernel functions can be employed to re-describe them in another higher dimensional space where they will be linearly separable, thus improving classification performance. More details can be found in [63,64,[67][68][69][70]. The principle of SVR is also based on finding the best hyperplane. ...
The study of the styrene–Ground Tire Rubber (GTR) graft radical polymerization is particularly challenging due to the complexity of the underlying kinetic mechanisms and nature of GTR. In this work, an experimental study on two scales (∼10 mL and ∼100 mL) and a machine learning (ML) modeling approach are combined to establish a quantitative relationship between operating conditions and styrene conversion. The two-scale experimental approach enables to verify the impact of upscaling on thermal and mixing effects that are particularly important in this heterogeneous system, as also evidenced in previous works. The adopted experimental setups are designed in view of multiple data production, while paying specific attention in data reliability by eliminating the uncertainty related to sampling for analyses. At the same time, all the potential sources of uncertainty, such as the mass loss along the different steps of the process and the precision of the experimental equipment, are also carefully identified and monitored. The experimental results on both scales validate previously observed effects of GTR, benzoyl peroxide initiator and temperature on styrene conversion but, at the same time, reveal the need of an efficient design of the experimental procedure in terms of mixing and of monitoring uncertainties. Subsequently, the most reliable experimental data (i.e., 69 data from the 10 mL system) are used for the screening of a series of diverse supervised-learning regression ML models and the optimization of the hyperparameters of the best-performing ones. These are gradient boosting, multilayer perceptrons and random forest with, respectively, a test R2 of 0.91 ± 0.04, 0.90 ± 0.04 and 0.89 ± 0.05. Finally, the effect of additional parameters, such as the scaling method, the number of folds and the random partitioning of data in the train/test splits, as well as the integration of the experimental uncertainties in the learning procedure, are exploited as means to improve the performance of the developed models.
... On the contrary, the ridge regression, every data is influential to estimation of parameters (Welling, 2004). In addition, the SVR can be flexible even in non-linear problems by utilizing kernel functions (Auria & Moro, 2008). ...
Regression analysis is one of the most widely utilized methods because of its Regression analysis is one of the most widely utilized methods because of its adaptability and simplicity. Recently, the machine learning (ML) approach, which adaptability and simplicity. Recently, the machine learning (ML) approach, which is one aspect of regression methods, has been gaining attention from researchers, is one aspect of regression methods, has been gaining attention from researchers, including social science, but there are only a few studies that compared the including social science, but there are only a few studies that compared the traditional approaches with the ML approach. This study was conducted to traditional approaches with the ML approach. This study was conducted to explore the usefulness of the ML approach by comparing the ordinary least square explore the usefulness of the ML approach by comparing the ordinary least square estimate (OLS), the stochastic gradient descent algorithm (SGD), and the support estimate (OLS), the stochastic gradient descent algorithm (SGD), and the support vector regression (SVR) with a model predicting and explaining the tuberculosis vector regression (SVR) with a model predicting and explaining the tuberculosis screening intention. The optimized models were evaluated by four aspects: screening intention. The optimized models were evaluated by four aspects: computational speed, effect and importance of individual predictor, and model computational speed, effect and importance of individual predictor, and model performance. The result demonstrated that each model yielded a similar direction performance. The result demonstrated that each model yielded a similar direction of effect and importance in each predictor, and the SVR with the radial kernel had of effect and importance in each predictor, and the SVR with the radial kernel had the finest model performance compared to its computational speed. Finally, this the finest model performance compared to its computational speed. Finally, this study discussed the usefulness and attentive points of the ML approach when a study discussed the usefulness and attentive points of the ML approach when a researcher utilizes it in the field of communication. researcher utilizes it in the field of communication.
... Like other regression issues, it is assumed that the relation between independent and dependent variables with algebraic function f(x), plus some noise (allowed error noise). For more informa tio n about SVR model and its structure, see Welling (2004), Awad and Khanna (2015) and Raji et al. (2022). The SVR model studied in this study is nonlinear that has three parameters. ...
Downscaling and simulating various meteorological variables at different time scales are fundamental topics for making climate change studies in a geographic region. Here, a new approach for downscaling the mean daily temperature was implemented using a vine copula‐based approach and considering the best CanESM2 predictors. The accuracy of the copula‐based approach was compared with genetic programming (GP), optimized support vector regression (OSVR), support vector machine (SVM), adaptive neuro‐fuzzy inference system (ANFIS) and artificial neural network (ANN) models at Birjand synoptic station in Iran. In the proposed approach, after examining the different vine copulas, the D‐vine copula was selected as the best copula according to the evaluation statistics and tree sequences. According to the root‐mean‐square error (RMSE) and Nash–Sutcliff efficiency (NSE), the accuracy of the ANN model in downscaling the mean daily temperature data was not acceptable and the other considered models were slightly overestimated. The results indicated that the copula‐based approach outperformed the other models in downscaling the mean daily temperature with NSE = 0.61. However, given the 99% confidence interval of the simulations, a slightly overestimation at temperatures above 20°C was observed for the copula‐based approach, which has better performance than the other considered models. The copula‐based approach was able to reduce RMSE by about 82, 20, 24, 47 and 34% compared to ANN, OSVR, GP, SVM and ANFIS models, respectively. The results also showed that the performance of the support vector regression model optimized by the ant colony algorithm is also acceptable and is in the second rank after the copula‐based approach. The accuracy of the copula‐based approach was also confirmed according to Taylor diagram and violin plot. The proposed approach has a higher accuracy than data‐driven models due to use of the conditional density of vine copulas, and the joint distribution of the mean daily temperature and selected predictors.
... The model is computationally light, and in some relatively simple cases can produce satisfactory results. Support Vector Machines (SVMs) and Support Vector Regression (SVR) [44] are well-known machine learning algorithms for solving classification and regression problems respectively. Over the years, they have been continuously gaining in popularity and adoption, for various applications, since they offer convenient implementation and testing facilities. ...
This article introduces a novel machine learning approach dedicated to the prediction of the bus arrival times in the bus stations over a given itinerary, based on the so-called Traffic Density Matrix (TDM). The TDM constructs a localized representation of the traffic information in a given urban area that can be specified by the user. We notably show the necessity of disposing of such data for successful, both short-term and long-term prediction objectives, and demonstrate that a global prediction approach cannot be a feasible solution. Several different prediction approaches are then proposed and experimentally evaluated on various simulation scenarios. They include traditional machine learning techniques, such as linear regression and support vector machines (SVM), but also advanced, highly non-linear neural network-based approaches. Within this context, various network architectures are retained and evaluated, including fully connected neural networks (FNN), convolutional neural networks (CNN), recurrent neural networks (RNN) and LSTM (Long Short Term Memory) approaches. The experimental evaluation is carried out under two types of different scenarios, corresponding to both long term and short-term predictions. To this purpose, two different data models are constructed, so-called ODM (Operator Data Model) and CDM (Client Data Model), respectively dedicated to long term and short-term predictions. The experimental results obtained show that increasing the degree of non-linearity of the predictors is highly benefic for the accuracy of the obtained predictions. They also show that significant improvements can be achieved over state of the art techniques. In the case of long-term prediction, the FNN method performs the best when compared with the baseline OLS technique, with a significant increase in accuracy (more than 66%). For short-term prediction, the FNN method is also the best performer, with more than 15% of gain in accuracy with respect to OLS.
... For prediction purposes, we have retained the Scikit-learn machine learning toolbox and tested two different approaches, a first one based on linear regression [17] and a second one exploiting SVR (Support Vector Regression) [18]. ...
... (SVR) Support Vector Machine (SVM) [6][7] is a supervised machine learning algorithm mostly used in classification, but easily adopted for regression since it form a generalization of the classification problem, in which the model returns a continuous-valued prediction. SVR can be defined as an optimization problem that relied on defining a convex ε-insensitive loss function, then trying to minimize this function and find the flattest tube that contains most of the training instances. ...
The overall equipment effectiveness (OEE) is a performance measurement metric widely used. Its calculation provides to the managers the possibility to identify the main losses that reduce the machine effectiveness and then take the necessary decisions in order to improve the situation. However, this calculation is done a-posterior which is often too late. In the present research, we implemented different Machine Learning algorithms namely; Support vector machine, Optimized Support vector Machine (using Genetic Algorithm), Random Forest, XGBoost and Deep Learning to predict the estimate OEE value. The data used to train our models was provided by an automotive cable production industry. The results show that the Deep Learning and Random Forest are more accurate and present better performance for the prediction of the overall equipment effectiveness in our case study.
... (SVR) Support Vector Machine (SVM) [6][7] is a supervised machine learning algorithm mostly used in classification, but easily adopted for regression since it form a generalization of the classification problem, in which the model returns a continuous-valued prediction. SVR can be defined as an optimization problem that relied on defining a convex ε-insensitive loss function, then trying to minimize this function and find the flattest tube that contains most of the training instances. ...
The overall equipment effectiveness (OEE) is a performance measurement metric wildly used. Its calculation provides to the managers the possibility to identify the main losses that reduce the machine effectiveness and then take the necessary decisions in order to improve the situation. However, this calculation is done a-posterior which is often too late. In the present research, we implemented different Machine Learning algorithms namely; Support vector machine, Optimized Support vector Machine (using Genetic Algorithm), Random Forest, XGBoost and Deep Learning to predict the estimate OEE value. The data used to train our models was provided by an automotive cable production industry. The results show that the Deep Learning and Random Forest are more accurate and present better performance for the prediction of the overall equipment effectiveness in our case study.
... First of all, because the output is a real number it becomes very difficult to predict the information at hand, which has infinite possibilities. In the case of regression, a margin of tolerance is set in approximation to the SVM which would have already requested from the problem (Parrella, 2007). Although the margin of tolerance is small, although it improves the regression and it increases the computational complexity. ...
The accurate prediction of lithium-ion battery Remaining Useful Life (RUL) is indispensable for safe and lifetime-optimised operation. Thereby, the monitoring of this vital component is very necessary for planning repair work and minimising unexpected electricity outage. However, the study and the investigation of internal battery parameters show several value changes within the battery lifetime and it is highly influenced by environmental and load conditions. Consequently, this paper presents a new prognostic method for online battery monitoring based on isometric feature mapping technique (ISOMAP) and incremental support vector regression (ISVR). ISOMAP is used to reduce some features extracted from lithium-ion batteries, with different health states, in both modes of charge and discharge, and ISVR is used to regress online the selected feature. Experimental results show that the proposed methodology provides a new suitable trend parameter for battery RUL prediction. Reference to this paper should be made as follows: Ben Ali, J. and Saidi, L. (2018) 'A new suitable feature selection and regression procedure for lithium-ion battery prognostics',
... In this method, we use Online SVR [9] as the learner. Moreover, we applied the RBF kernel [13] as the kernel function for Online SVR. ...
This study investigates the effectiveness of reduction of training sets and kernel space for action decision using future prediction. For future prediction in a real environment, it is necessary to know the properties of the state and disturbance resulting from the outside environment, such as a ground surface or water surface. However, obtaining the properties of the disturbance depends on the specifications of the target processor,especially its sensor resolution or processing ability. Therefore, sampling-rate settings are limited by hardware specifications. In contrast, in the case of future prediction using machine learning, prediction is based on the tendency obtained from past training or learning. In such a situation, the learning time is proportional to training data. At worst, the prediction algorithm will be difficult to implement in real time because of time complexity. Here, we consider the possibility of carefully analyzing the algorithm and applying dimensionality reduction technique st accelerate the algorithm. In particular, to reduce the training sets and kernel space based on the recent tendency of disturbance or state, we focus on the use of the fast Fourier transform (FFT) and pattern matching. From this standpoint, we propose a method for dynamically reducing the dimensionality based on the tendency of disturbance. As a future application, an algorithm for operating unmanned agricultural support machines will be planned to implement the proposed method in a real environment.
... In this method, we use Online SVR [9] as a learner. Moreover, we applied RBF kernel [13] as the kernel function to the Online SVR of the learner. ...
Considering a working in a real environment based on future prediction, it’s necessary to know the property of its state and disturbance that will be given by the outside environment. On the other hand, obtaining the property of the disturbance depends on specification for target processor, especially, sensor resolution or processing ability of the processor. Therefore, sampling rate settings will be limited by hardware specification. In contrast, in case of a future prediction using a machine learning, it predicts that based on the tendency that obtained by past training or learning. In this kind of situation, the learning time will be proportionally larger to training data. At worst, the prediction algorithm will be hard to work in real time due to time-complexity. In the proposed method, the possibility of carefully analyzing the algorithm and applying dimensionality reduction techniques in order to accelerate the algorithm has been considered.
... As mentioned above, defines the error-insensitive tube around the regression function and thus controls how well the function fits the training data [20]. The parameter controls the tradeoff between training error and model complexity; a smaller increases the number of training errors; a larger increases the penalty for training errors and results in a behavior similar to that of a hard-margin SVM [21]. ...
This paper presents a model for very short-term load forecasting (VSTLF) based on algebraic prediction (AP) using a modified concept of the Hankel rank of a sequence. Moreover, AP is coupled with support vector regression (SVR) to accommodate weather forecast parameters for improved accuracy of a longer prediction horizon; thus, a hybrid model is also proposed. To increase system reliability during peak hours, this prediction model also aims to provide more accurate peak-loading conditions when considerable changes in temperature and humidity happen. The objective of going hybrid is to estimate an increase or decrease on the expected peak load demand by presenting the total MW per Celsius degree change (MW/C°) as criterion for providing a warning signal to system operators to prepare necessary storage facilities and sufficient reserve capacities if urgently needed by the system. The prediction model is applied using actual 2014 load demand of mainland South Korea during the summer months of July to September to demonstrate the performance of the proposed prediction model.
... The resulting SVM takes the following form (Smola and Vapnik, 1997;Welling, 2004). ...
Wellhead chokes are widely used in the petroleum industry. Owning to the high sensitivity of oil and gas production to choke size, an accurate correlation to specify choke performance is vitally important. The aim of this contribution was to develop effective relationships among the liquid flow rate, gas liquid ratio, flowing wellhead pressure, and surface wellhead choke size using the support vector machines (SVMs). The accurate data set was gathered from the 15 different fields containing 100 production samples from the vertical wells at wide ranges of wellhead choke sizes. This computational model was compared with the previous developed correlations in order to investigate its applicability for subcritical two phase flow regimes through wellhead chokes. Results confirmed amazing capability of the SVM to predict liquid flow rates. The value of R² obtained was 0.9998 for the SVM model. This developed predictive tool can be of massive value for petroleum engineer to have accurate estimations of liquid flow rates through wellhead chocks.
... I , y L )} 4 (X Â Y) l ) to not exceed the maximum amount 3 and thus made it as at as could reasonably be expected. 37,38 One of the most critical concepts in SVR cases is that exhibiting the solution by means of a small subset of training points gives huge computational benets. In the SVR algorithm, the training dataset contained convex optimization wherein there were no local minima. ...
Nowadays the detection of proteins plays a crucial role for the early diagnosis of diseases. The combination of biosensor application with nanotechnology has offered new alternatives for clinical diagnostic techniques. One of the major public health problems in many developing countries is tuberculosis (TB) susceptibility and Interferon-gamma (IFN-γ) can be used in the diagnosis of this infectious disease. In this study, a prototype graphene based FET structure was employed as a biosensor. Additionally, a PDMS layer was deployed beneath the graphene as a dielectric layer. As a result of the changeability of Ids (drain-source current), the carrier concentration would change when the IFN-γ molecules attach to the surface of graphene. To acquire another pattern for the I-V (current-voltage characteristic), the Incremental Support Vector Regression (ISVR) algorithm was also employed. The comparative study based on the outcomes of the ISVR and pre-existing analytical models with experimental data found that there was acceptable agreement, which was able to substantiate the proposed models. Moreover, the ISVR showed that the proposed method remarkably improved the accuracy of prediction.
... . These direct approaches include: neural networks [71], support vector regression (SVR) [65][66][67], and kernel adaptive filter algorithms such as kernel recursive least squares (KRLS) [16] and kernel least mean squares (KLMS) [15]. These approaches usually have good performance to predict the noiseless time series. ...
There are numerous dynamical system applications that require estimation or prediction from noisy data, including vehicle tracking, channel tracking, time series denoising, prediction, estimation, and so on. Many linear algorithms have been developed to deal with these problems under different assumptions and approximations, such as the Kalman filter, the recursive least squares algorithm, and the least mean squares algorithm. However, these linear algorithms cannot solve nonlinear problems that often occur in real life. To address these nonlinear problems, some nonlinear algorithms have been recently proposed, like kernelized version of the linear algorithms. Our research follows this line and seeks to develop novel algorithms using kernel methods to deal with nonlinear problems. Specifically, our goal is to derive the Kalman filter in the reproducing kernel Hilbert space (RKHS), which is a space of functions, to implement signal denoising, prediction and estimation.
... exceeding a maximum value e and flattens the dataset as much as possible (Stahlbock and Lessmann 2004;Welling 2004). In other words, errors are neglected as long as they are smaller than a pre-specified value e. ...
This paper reports on a set of experiments designed to develop a workable gas sensor prototype using an electronic system with methane. The current is found to be sensitive to the presence of methane gas,
which is a conduit for a variety of gas sensors. The sensitivity is shown to depend on pointed or broad electrode configurations. Scanning electron microscopy images show the area of conductance that determines the quality of the electrodes in three configurations. Data processing is performed with a support vector regression algorithm in conjunction with statistical analysis for error and quality control. The reported results can be adapted to a broad range of industrial applications for enhanced productivity, safety, innovation, data processing, and overall total quality management.
... Support vector machines (SVM), which were developed by Vapnik (1995) as a tool for classification and regression, embody the structural risk minimization principle, unlike conventional neural networks which adhere to the empirical risk minimization principle (Vapnik, 1995). All SVR models were created using the OnlineSVR software created by Parrella (2007), which can be used to build support vector machines for regression. The data was partitioned into two sets: a calibration set and a validation set. ...
This study explores the ability of machine learning techniques coupled with the bootstrap method to forecast drought conditions. The two machine learning techniques that were used were artificial neural networks and support vector regression. The potential of the bootstrap technique to reduce the uncertainty and develop reliable artificial neural network (ANN) and support vector regression (SVR) models was explored. The performance of the bootstrap-ANN (BANN) and bootstrap-SVR (BSVR) models were compared to traditional ANN and SVR models that do not use the bootstrap method. The Standard Precipitation Index (SPI) (in this case SPI 3, SPI 12 and SPI 24) was the drought index that was forecast using the aforementioned models. These SPI values represent short and long-term drought conditions and are analogous to agricultural and hydrological drought conditions, respectively. The performances of all models were compared using RMSE, MAE, R2 and a measure of persistence. It was determined that the use of the bootstrap method reduced forecast uncertainties and improved the performance of the machine learning methods.
... In recent decades, kernel methods have been widely applied in classification and regression, such as the support vector machine (SVM) [1] and support vector regression (SVR) [2]. The basic idea behind kernel methods is that a positive definite kernel function defined on pairs of input vectors can map a vector in the input space into a higher dimensional (or infinite dimensional) feature space (often called a reproducing kernel Hilbert space) and the inner product in the feature space can be computed using the kernel function, instead of the high dimensional inner product. ...
In this paper, the extended kernel recursive least squares (Ex-KRLS) algorithm is reviewed and analyzed. We point out that the Theorem 1 in [10] is not always correct for general cases. Furthermore, the Ex-KRLS algorithm for tracking model is just a random walk KRLS algorithm. Finally, this algorithm is explained as a special Kalman filter in the reproducing kernel Hilbert space.
... .ðx I ; y L Þg ðX Â YÞ l Þ not to exceed a maximum value e which makes it as flat as possible (Stahlbock and Lessmann 2004;Welling 2004). In other words, errors are neglected as long as they are smaller than a prespecified value e. ...
This experimental study attempts to design and develop a novel carbon-based material fabrication method that implements high voltage AC arc discharge. The carbon-based material is to be used in an electric circuit to build up a gas sensing mechanism. After applying a high AC voltage to the graphite electrodes employed in the experiment, an arc ignites between the electrodes as the gas between them is iodized. In the carbon decomposition process, the arc is employed and results in the fabricated carbon strands. Thereafter, carbon monoxide (CO2) in the atmospheric pressure is passed over the electrodes in the Pyrex glass tube chamber where the carbon strand fabrication process takes place. An increment appears in the measured current voltage (I–V) between the two sides of the carbon strand as the carrier concentration of CO2 gas is increased from 200 to 800 ppm. Support vector regression (SVR) algorithm was used for data processing with statistical analysis for errors and quality control.
... This configuration focuses on checking the system response and finding the kinetics of bacteria binding. The sensor was kept in chambers with 10 4 cfu/mL of E. coli [20,37]. As higher numbers of E. coli are caught by the graphene film antibodies, the conductance of the channel increases. ...
In this research, a set of novel models based on field effect transistor (FET) structure using graphene have been proposed with the current–voltage (I–V) characteristics of graphene employed to model the sensing mechanism. It has been observed that the graphene device experiences a drastic increase in conductance when exposed to Escherichia coli bacteria at 0– (Formula presented.) cfu/mL concentrations. Hence, simplicity of the structure, fast response rate and high sensitivity of this nanoelectronic biosensor make it a more suitable device in screening and functional studies of antibacterial drugs and an ideal high-throughput platform that can detect any pathogenic bacteria. Accordingly, the proposed model exhibits a satisfactory agreement with the experimental data.
... ðx I ,y L Þg ðX Â YÞ l Þ not to exceed a maximum value " and makes it as flat as possible. 40,41 In other words, errors are neglected as long as they are smaller than a pre-specified value ". Unlike the ANN, training of the SVR includes convex optimization in which there are no local minima. ...
Graphene is another allotrope of carbon with two-dimensional monolayer honeycomb. Owing to its special characteristics including electrical, physical and optical properties, graphene is known as a more suitable candidate compared to other materials to be used in the sensor application. It is possible, moreover, to use biosensor by using electrolyte-gated field effect transistor based on graphene (GFET) to identify the alterations in charged lipid membrane properties. The current article aims to show how thickness and charges of a membrane electric can result in a monolayer graphene-based GFET while the emphasis is on the conductance variation. It is proposed that the thickness and electric charge of the lipid bilayer (LLP and QLP) are functions of carrier density, and to find the equation relating these suitable control parameters are introduced. Artificial neural network algorithm as well as support vector regression has also been incorporated to obtain other models for conductance characteristic. The results comparison between analytical models, artificial neural network and support vector regression with the experimental data extracted from previous work show an acceptable agreement.
© The Author(s) 2015 Reprints and permissions: sagepub.co.uk/journalsPermissions.nav.
... Detailed descriptions of SVR model development can be found in Cimen (2008). All SVR models were created using the OnlineSVR software created by Parrella (2007), which can be used to build support vector machines for regression. The data was partitioned into two sets: a calibration set and a validation set. ...
... For the autonomous model in Fig. 1, or the embedding past measurements are used to predict , by approximating the function or , where and are the embedding dimension and the embedding delay, respectively, according to Takens' embedding theorem [1]. These direct approaches include: neural networks [2], support vector regression (SVR) [3], [4], and kernel adaptive filter algorithms such as kernel recursive least squares (KRLS) [5] and kernel least mean squares (KLMS) [6]. These approaches usually have good performance to predict the noiseless time series. ...
This paper presents a novel generative model for time series based on the Kalman filter algorithm in a reproducing kernel Hilbert space (RKHS) using the conditional embedding operator. The end result is a nonlinear model that quantifies the hidden state uncertainty and propagates its probability distribution forward as in the Kalman algorithm. The embedded dynamics can be described by the estimated conditional embedding operator constructed directly from the training measurement data. Using this operator as the counterpart of the state transition matrix, we reformulate the Kalman filter algorithm in RKHS. For the state model, the hidden states are the estimated embeddings of the measurement distribution, while the measurement model serves to connect the estimated measurement embeddings with the current mapped measurements in the RKHS. This novel algorithm is applied to noisy time-series estimation and prediction, and simulation results show that it outperforms other existing algorithms. In addition, improvements are proposed to reduce the size of the operator and reduce the computation complexity. Index Terms—Kalman filter, RKHS, KRLS, conditional embedding operator.
... SVR models were developed using the OnlineSVR software created by Parrella [59]. OnlineSVR is a technique used to build support vector machines for regression. ...
Drought forecasts can be an effective tool for mitigating some of the more adverse consequences of drought. Data-driven models are suitable forecasting tools due to their rapid development times, as well as minimal information requirements compared to the information required for physically based models. This study compares the effectiveness of three data-driven models for forecasting drought conditions in the Awash River Basin of Ethiopia. The Standard Precipitation Index (SPI) is forecast and compared using artificial neural networks (ANNs), support vector regression (SVR), and wavelet neural networks (WN). SPI 3 and SPI 12 were the SPI values that were forecasted. These SPI values were forecast over lead times of 1 and 6 months. The performance of all the models was compared using RMSE, MAE, and R 2 . The forecast results indicate that the coupled wavelet neural network (WN) models were the best models for forecasting SPI values over multiple lead times in the Awash River Basin in Ethiopia.
... c) Global variables are defined as P 1 and C, among which P 1 is the width of kernel function, while C is a coefficient. Both of them are given idiographic data according to different models [12], [20], [21]. d) A kernel-function wavelet and insensitive coefficient are chosen in this process for model training and testing through MATLAB Version 7.5. ...
The prediction algorithm is one of the most important factors in the quality of wind-power prediction. In this paper, based on the principles of wavelet transform and support vector machines (SVMs), as well as the characteristics of wind-turbine generation systems, two prediction methods are presented and discussed. In method 1, the time series of model input are decomposed into different frequency modes, and the models are set up separately based on the SVM theory. The results are combined together to forecast the final wind-power output. For comparison purposes, the wavelet kernel function is applied in place of the radial basis function (RBF) kernel function during SVM training in method 2. The operation data of one wind farm from Texas are used. Mean relative error and relative mean square error are used to evaluate the forecasting errors of the two proposed methods and the RBF SVM model. The means of evaluating the prediction-algorithm precision is also proposed.
... The method of Support Vector Regression (SVR) is based on statistical learning theory by Vapnik [8] T the functional form as flat as possible [4,9,10]. SVR is formulated as convex optimization problem with slack variables ξ i ,ξ i * to allow for model errors [11,12] ( ) (1) which controls the trade-off between overfitting and model complexity through a regularization parameter C>0 [8]. We employ ε-SVR, using an ε-intensive loss function, that assigns an error only to those observations ξ i ,ξ i * ≥ 0 outside the ε-insensitive tube [13,14], named support vectors [15], using the loss function ...
Support Vector Regression is one of the promising contenders in predicting the 111 time series of the NN3 Neural Forecasting Competition. As they offer substantial degrees of freedom in the modeling process, in selecting the kernel function and its parameters, cost and epsilon parameters, issues of model parameterization and model selection arise. In lack of an established methodology or comprehensive empirical evidence on their modeling, a number of heuristics and ad-hoc rules have emerged, that result in selecting different models, which show different performance. In order to determine a lower bound for Support Vector Regression accuracy in the NN3 competition, this paper seeks to compute benchmark results using a naive methodology with a fixed parameter grid-search and exponentially increasing step sizes for radial basis function kernels, estimating 43,725 candidate models for each of the 111 time series. The naive approach attempts to mimic many of the common mistakes in model building, providing error as a lower bound to support vector regression accuracy. The in-sample results parameters are evaluated to estimate the impact of potential shortcomings in the grid search heuristic and the interaction effects of the parameters.
Plant biomass estimation is critical due to the variability of different environmental factors and crop management practices associated with it. The assessment is largely impacted by the accurate prediction of different environmental sustainability indicators. A robust model to predict sustainability indicators is a must for the biomass community. This study proposes a robust model for biomass sustainability prediction by analyzing sustainability indicators using machine learning models. The prospect of ensemble learning was also investigated to analyze the regression problem. All experiments were carried out on a crop residue data from the Ohio state. Ten machine learning models, namely, linear regression, ridge regression, multilayer perceptron, k-nearest neighbors, support vector machine, decision tree, gradient boosting, random forest, stacking and voting, were analyzed to estimate three biomass sustainability indicators, namely soil erosion factor, soil conditioning index, and organic matter factor. The performance of the model was assessed using cross-correlation (R2), root mean squared error and mean absolute error metrics. The results showed that Random Forest was the best performing model to assess sustainability indicators. The analyzed model can now serve as a guide for assessing sustainability indicators in real time.
Exploring the data-driven prediction strategy of physical and chemical properties is attractive for the rational design of crystal dielectrics with target characteristics, especially for the optoelectronics and electronic devices. To accelerate the design and discovery of novel crystal based dielectric devices, this work proposed an interpretable hybrid machine learning (ML) model to predict the dielectric constant of halide alkali crystals. The approach uses Particle Swarm Optimization-Random Forest (PSO-RF) model and polarization ion parameters of molecular dynamics simulation accurately predict the dielectric constant of 19 alkali halides on a small data set. The predicted performance of the PSO-RF model reaches R² of 0.987, which exceeds other seven popular machine learning models. The PSO-RF model is interpreted by feature importance and partial dependence plots. This work allows for data expansion using experimental data and molecular structure information, provides ideas for predicting the physical and chemical properties of materials on small data sets.
Recently, there has been an increasing interest in the analysis of flow cytometry data, which involves measurements of a set of surface and functional markers across hundreds and thousands of cells. These measurements can often be used to differentiate various cell types and there has been a rapid development of analytic approaches for achieving this. However, in spite of the fact that measurements are available on such a large number of cells, there have been very limited advances in deep learning approaches for the analysis of flow cytometry data. Some preliminary work has focused on using deep learning techniques to classify cell types based on the cell protein measurements. In a first of its kind study, we propose a novel deep learning architecture for predicting functional markers in the cells given data on surface markers. Such an approach is expected to automate the measurement of functional markers across cell samples, provided data on the surface markers are available, that has important practical advantages. We validate and compare our approach with competing machine learning methods using a real flow cytometry dataset, and showcase the improved prediction performance of the deep learning architecture.
In this work our aim was to use support vector regression technique to predict unemployment rate and gross domestic product (GDP) of the Organization for the Economic Cooperation and Development (OECD) countries. The original data for unemployment rate and GDP taken from OECD was tested with the models generated by support vector regression. The analysis of using different kernels as well as analysis of using extended data from four countries in predicting the unemployment rate and GDP were considered separately. For unemployment rate, the relative error in average was 2.055% using the radial kernel and using extended data from four countries. Similarly, for GDP the relative error in average was 0.503%. It was observed that in fact support vector regression is a very powerful modelling tool in using the existing historical data for predicting the values of interest.
Over recent years, a lot of money have been spent by the oil and gas industry to maintain pipeline integrity, specifically in handling internal corrosion. In fact, current solutions in pipeline corrosion maintenance are extremely costly to the companies. The empirical solutions also lack intelligence in adapting to different environment. In the absence of a suitable algorithm, the time taken to determine the corrosion occurrence is lengthy as a lot of testing is needed to choose the right solution. If the corrosion failed to be determined at an early stage, the pipes will burst leading to high catastrophe for the company in terms of costs and environmental effect. This creates a demand of utilizing machine learning in predicting corrosion occurrence. This paper discusses on the evaluation of machine learning algorithms in predicting internal corrosion rate. It is because there are still gaps on study on evaluating suitable machine learning algorithms for corrosion prediction. The selected algorithms for this paper are Artificial Neural Network, Support Vector Machine and Random Forest. As there is limited data available for corrosion studies, a synthetic data was generated. The synthetic dataset was generated via random Gaussian function and incorporated de Waard-Milliams model, an empirical determination model for internal corrosion. Based on the experiment conducted, Artificial Neural Network shows a more robust result in comparison to the other algorithms.
Machine learning requires sufficient and reliable data to enhance the prediction performance. However, environmental data sometimes is short and/or contains missing data. Often existing prediction models built on machine learning fail to predict environmental problems accurately. We argue that spatial domain data can be used to facilitate the training of temporal prediction model. This paper formulates mathematically a spatial data aided incremental support vector regression (SalncSVR) for spatio-temporal PM2.5 prediction. We conduct spatio-temporal PM2.5 prediction over 13 monitoring stations in Auckland New Zealand, and compare the proposed SalncSVR with a pure temporal IncSVR prediction.
Droughts can cause significant damage to agricultural and other systems. An important aspect of mitigating the impacts of drought is an effective method of forecasting future drought events. In this study, five methods of forecasting drought for short lead times were explored in the Awash River Basin of Ethiopia. The Standard Precipitation Index (SPI) was the drought index chosen to represent drought in the basin. Machine learning techniques including artificial neural networks (ANNs) and support vector regression (SVR) were compared with coupled models (WA-ANN and WA-SVR) which pre-process input data using wavelet analysis (WA). This study proposed and tested the SVR and WA-SVR methods for short term drought forecasting. This study also used only the approximation series (derived via wavelet analysis) as inputs to the ANN and SVR models, and found that using just the approximation series as inputs for models gave good forecast results. The forecast results of all five data driven models were compared using several performance measures (RMSE, MAE, R
2 and a measure of persistence). The forecast results of this study indicate that the coupled wavelet neural network (WA-ANN) models were the best models for forecasting SPI 3 and SPI 6 values over lead times of 1 and 3 months in the Awash River Basin.
Klinkenberg permeability is an important parameter in tight gas reservoirs. There are conventional methods for determining it, but these methods depend on core permeability. Cores are few in number, but well logs are usually accessible for all wells and provide continuous information. In this regard, regression methods have been used to achieve reliable relations between log readings and Klinkenberg permeability. In this work, multiple linear regression, tree boost, general regression neural network, and support vector machines have been used to predict the Klinkenberg permeability of Mesaverde tight gas sandstones located in Washakie basin. The results show that all the four methods have the acceptable capability to predict Klinkenberg permeability, but support vector machine models exhibit better results. The errors of models were measured by calculating three error indexes, namely the correlation coefficient, the average absolute error, and the standard error of the mean. The analyses of errors show that support vector machine models perform better than the other models, but there are some exceptions. Support vector machine is a relatively new intelligence method with great capabilities in regression and classification tasks. Herein, support vector machine was used to predict the Klinkenberg permeability of a tight gas reservoir and the performances of four regression techniques were compared.
As outlined in section 1.1, there will be an inevitable difference between the current target position and the position of the treatment beam. This difference originates from the latency in acquiring and computing the target position and from moving the robot. The goal clearly is to compensate for this delay and try to send the robot to a predicted position. In practice, several methods for the compensation of this delay are implemented and used clinically. These are a pattern-matching algorithm, called Zero Error Prediction (ZEP), an adaptive filter based on Least Mean Squares (LMS), a fuzzy prediction algorithm and a hybrid combination of these [35, 36].
A data-based adaptive online prediction model is proposed for plant-wide production indices based on support vector regression, a general method which we customized specifically to model very large data sets that are generated dynamically and periodically. The proposed model can update its parameters online according to the statistical properties of the training samples. Further, in order to improve the prediction precision, each sample is weighted with a dynamic penalty factor that considers the effect of each sample on the prediction model accuracy. Moreover, a customized procedure is introduced to handle large training sets. After having been convincingly evaluated on benchmark data, effectiveness and performance of our approach for plant-wide production indices is demonstrated using industrial data from an operating ore dressing plant over a range of scale in training data set size. The higher accuracy and shorter computation times than existing methods suggest that it may prove advantageous in actual application to dynamic production processes.
Soft sensors have been widely used in chemical plants to estimate process variables that are difficult to measure online. One of the crucial difficulties of soft sensors is that predictive accuracy drops due to changes in state of chemical plants. Characteristics of adaptive soft sensor models such as moving window models, just-in-time models and time difference models were previously discussed. The predictive accuracy of any traditional models decreases when sudden changes in processes occur. Therefore, a new soft sensor method based on online support vector regression (SVR) and the time variable was developed for constructing soft sensor models adaptive to rapid changes of relationships among process variables. A nonlinear SVR model with the time variable is updated with the most recent data. The proposed method was applied to simulation data and real industrial data, and achieved higher predictive accuracy than traditional ones even when time-varying changes in process characteristics happen. © 2013 American Institute of Chemical Engineers AIChE J, 2013
The simultaneous identification of multiple nonlinear plants using support vector regressions, and the methodology of multiple models, switching, and tuning, is proposed in this paper. Models are randomly initialised and, at each instant, they are assigned to a specific nonlinear plant, and adapted toward it. Accordingly, the models self-organise themselves in a manner so that, at any given instant of time, a plant is tracked by only one model. Two model assignment strategies are considered in this paper. The choice of the assignment strategy and the performance criteria play an important role in the model assignment, and hence the convergence.
In this paper we propose a new classifier called an incremental learning algorithm based on support vector machine with Mahalanobis distance (ISVMM). Prediction of the incoming data type by supervised learning of support vector machine (SVM), reducing the step of calculation and complexity of the algorithm by finding a support set, error set and remaining set, providing of hard and soft decisions, saving the time for repeatedly training the datasets by applying the incremental learning, a new approach for building an ellipsoidal kernel for multidimensional data instead of a sphere kernel by using Mahalanobis distance, and the concept of handling the covariance matrix from dividing by zero are various features of this new algorithm. To evaluate the classification performance of the algorithm, it was applied on intrusion prevention by employing the data from the third international knowledge discovery and data mining tools competition (KDDcup'99). According to the experimental results, ISVMM can predict well on all of the 41 features of incoming datasets without even reducing the enlarged dimensions and it can compete with the similar algorithm which uses a Euclidean measurement at the kernel distance.
Many real scenarios in machine learning are non-stationary. These challenges forces to develop new algorithms that are able to deal with changes in the underlying problem to be learnt. These changes can be gradual or abrupt. As the dynamics of the changes can be different, the existing machine learning algorithms exhibit difficulties to cope with them. In this work we propose a new method, that is based in the introduction of a forgetting function in an incremental online learning algorithm for two-layer feedforward neural networks. This forgetting function gives a monotonically crescent importance to new data. Due to this fact, the network forgets in presence of changes while maintaining a stable behavior when the context is stationary. The theoretical basis for the method is given and its performance is illustrated by evaluating its behavior. The results confirm that the proposed method is able to work in evolving environments.
ResearchGate has not been able to resolve any references for this publication.