Fig 5 - uploaded by Ricardo Cerri
Content may be subject to copyright.
Contexts in source publication
Context 1
... Nemenyi post-hoc test was then applied to identify which pairwise comparisons presented statistically significant differences. The critic diagram in Figure 5 shows the results of the Nemenyi test. We connected methods where no statistically significant results were detected. ...
Context 2
... to Figure 5, the only method which obtained statistically better results than SOM-MLL was the SVM with the Label Powerset transformation. We would like to emphasize, however, that this conclusion is based on the averaged f-measure values obtained considering the fmea- sure on all datasets. ...
Similar publications
Citations
... For example, the multi-scale attention-guided convolutional neural network is presented by Zheng et al. (2022) to effectively capture the varied properties of gastric cancer lesions. The authors demonstrated improved segmentation accuracy compared to traditional methods, highlighting the potential of advanced machine learning techniques for this task [15], [16], [17], [18] and Li et al. (2021) proposed a hybrid segmentation framework combining multiple approaches, including thresholding, region growing, and clustering, for the detection and delineation of gastric ulcers in endoscopic images. Their method addressed the challenges of varied lesion characteristics and showed promising results for clinical applications, yet the specificity for bleeding tissues in gastric images remains under-explored [19]. ...
Timely and precise classification and segmentation of gastric bleeding in endoscopic imagery are pivotal for the rapid diagnosis and intervention of gastric complications, which is critical in life-saving medical procedures. Traditional methods grapple with the challenge posed by the indistinguishable intensity values of bleeding tissues adjacent to other gastric structures. Our study seeks to revolutionize this domain by introducing a novel deep learning model, the Dual Spatial Kernelized Constrained Fuzzy C-Means (Deep DuS-KFCM) clustering algorithm. This Hybrid Neuro-Fuzzy system synergizes Neural Networks with Fuzzy Logic to offer a highly precise and efficient identification of bleeding regions. Implementing a two-fold coarse-to-fine strategy for segmentation, this model initially employs the Spatial Kernelized Fuzzy C-Means (SKFCM) algorithm enhanced with spatial intensity profiles and subsequently harnesses the state-of-the-art DeepLabv3+ with ResNet50 architecture to refine the segmentation output. Through extensive experiments across mainstream gastric bleeding and red spots datasets, our Deep DuS-KFCM model demonstrated unprecedented accuracy rates of 87.95%, coupled with a specificity of 96.33%, outperforming contemporary segmentation methods. The findings underscore the model's robustness against noise and its outstanding segmentation capabilities, particularly for identifying subtle bleeding symptoms, thereby presenting a significant leap forward in medical image processing.
... Then, when a test data is aimed to be clustered, it is mapped to the topographic map, which means BMU is defined for this data. Based on the BMU label, which is defined in the training process, the test data is clustered [27]. ...
Online monitoring of electric power components in smart grids is of great importance to enhance reliability. Fault detection at primary levels in distribution transformers, the chief components to maintaining the integrity of modern power networks, prevents following significant destructive damages and high costs of failures in smart grids. Data-driven structure in smart grids provides accessibility to data related to the condition of transformers in data centers. Frequency response analysis (FRA), an efficient and sensitive technique to identify transformer defects, can be utilized in online monitoring. However, a trustworthy and consistent code for interpreting frequency responses has not yet been proposed by standards. This study proposes a self-organizing map (SOM) neural network as an intelligent interpreter using appropriate feature groups obtained from suitable statistical indices (SIns). In order to distinguish the severities and locations of disk space variation (DSV) defects as common faults in transformers, an experimental setup including 20 kV windings of a 1.6 MVA distribution transformer and an impedance analyzer are provided. The promising performance of SOM in detecting DSV faults with 100% accuracy shows that the proposed method is capable of identifying faults using high dimensional and nonlinear FRA data sets.
... In unsupervised manner, Gustavo et al. [22] explore the power of SOM for multi-label classification. Since the SOM has ability to map input instances to a map of neurons. ...
... The process is continued as follows: the weights of each neighborhood is also updated, and approximate to the winning neuron; a good choice for finding for the neighborhood is using Gaussian function 6, where h j,i is the neighborhood of the winning neuron i, while j is the older winning neuron, the distance d j,i is a distance between neurons, the defines the spreading of neighborhoods. [22] ...
... Second, the winning neuron is selected from the neuron grid Ω using the distance metric (e.g., Euclidean). Finally, the weights of all related neurons will be updated using Eq. 8 [22]. The process is terminated when the network is converged, and the weights in the map might have the same distribution as the input vectors. ...
The autoencoder-based latent representations have been widely developed for unsupervised learning in cyber-security domain, and has shown remarkable performance. Our previous work has introduced a hybrid autoencoders (AEs) and self-organizing maps (SOMs) for unsupervised IoT malware detection. However, the paper has only examined the characteristics of the latent representation of ordinary AEs in comparison to that of principle component analysis (PCA) on various IoT malware scenarios. This paper extends the work by employing denoising AEs (DAEs) to enhance the generalization ability of latent representations as well as optimizing hyper-parameters of SOMs to improve the hybrid performance. Particularly, this aims to further examine the characteristics of AE-based structure models (i.e., DAE) for identifying unknown/new IoT attacks and transfer learning. Our model is evaluated and analyzed extensively in comparison with PCA and AEs by a number of experiments on the NBaIoT dataset. The experimental results demonstrate that the latent representation of DAEs is often superior to that of AEs and PCAs in the task of identifying IoT malware.
... After extracting the delta-radiomics features from the original ROI, due to the large number of radiomic features extracted from the images, many methods mentioned above are used to rule out redundant delta-radiomic features (DRFs). The selected DRFs are then tested to determine their significance as a treatment response function using linear regression models, t-test, and mixed-effect models (50). Significant DRFs are further tested and modeled using machine-learning algorithms to create a model that can predict the outcome of a new patient. ...
By breaking the traditional medical image analysis framework, precision medicine–radiomics has attracted much attention in the past decade. The use of various mathematical algorithms offers radiomics the ability to extract vast amounts of detailed features from medical images for quantitative analysis and analyzes the confidential information related to the tumor in the image, which can establish valuable disease diagnosis and prognosis models to support personalized clinical decisions. This article summarizes the application of radiomics and dosiomics in radiation oncology. We focus on the application of radiomics in locally advanced rectal cancer and also summarize the latest research progress of dosiomics in radiation tumors to provide ideas for the treatment of future related diseases, especially ¹²⁵I CT-guided radioactive seed implant brachytherapy.
... In unsupervised manner, Gustavo et. al. [6] explore the power of SOM for multi-label classification. Since the SOM has ability to map input instances to a map of neurons. ...
... The process is continued as follow: the weights of each neighborhood also is updated, and approximate to the winning neuron; a good choice for finding for the neighborhood is using Gaussian function 4, where h j,i is the neighborhood of the winning neuron i, while j is the older winning neuron, the distance d j,i is a distance between neurons, the σ defines the spreading of neighborhoods. [6]. ...
... Finally, the weight vector at iteration (t + 1) is updated by Equation 6. ...
The feature representation of AutoEncoders (AEs) has been widely used for unsupervised learning, particularly in cybersecurity domain, and demonstrated promising performance. However, deeply investigations of the feature learner for the task of IoT attack detection in unsupervised learning have not been carried out yet. In this paper, we study the feature representation of AEs in combination with a subsequent clustering-based technique like Self-Organizing Maps (SOM) for unsupervised learning IoT attack detection. This aims to get insight into the characteristics of the AE learners in the tasks of unsupervised IoT detection such as identifying unknown/new IoT attacks and transfer learning. To highlight the behavior of AE-based learners, a feature reduction like Principle Component Analysis (PCA) is used to construct a feature space for facilitating SOM. The proposed models are investigated and assessed extensively by a number of experiments and analyses on the NBaIoT dataset. The experimental results highly suggest that AEs should be used for transferring models as training data is highly un-balanced and includes IoT attacks being similar to Benign. If the training data seems to be balanced, and contains IoT attacks being significantly deviated from Benign, the feature reduction like PCA is more preferable.
... Several studies have realized multi-label classification by using a clustering algorithm. A typical type of clusteringbased multi-label classification algorithm utilizes SOM [41,42]. Although the learning process of SOM is performed in an unsupervised learning manner, the convergence of the SOM network is significantly slow and unstable. ...
This paper proposes a multi-label classification algorithm capable of continual learning by applying an Adaptive Resonance Theory (ART)-based clustering algorithm and the Bayesian approach for label probability computation. The ART-based clustering algorithm adaptively and continually generates prototype nodes corresponding to given data, and the generated nodes are used as classifiers. The label probability computation independently counts the number of label appearances for each class and calculates the Bayesian probabilities. Thus, the label probability computation can cope with an increase in the number of labels. Experimental results with synthetic and real-world multi-label datasets show that the proposed algorithm has competitive classification performance to other well-known algorithms while realizing continual learning.
... • The inflexibility of existing models to analyze multiple academic and non-academic factors that are deemed to influence the quality of student learning. [11], [27], [41], [43], [45], [50], [51]) and their Research Gaps (denoted as RG). ...
... Much of these works are Binary-Relevance based [65] on which the labelled classes are tackled independently. Towards a more dependable generalization, however, one should rely on Label-Powerset (defined in [50]) and take the correlation between labeled classes into consideration. In our research, it is of great importance to recognize the relationship between gv i , which principally can be reached by clustering related sets of labels. ...
... The closely related works, in terms of using SOM for multi-label prediction, are [50], [51]. We are mostly in line with [51] in (1) determining the n win for a test input instance ... x i by minimising the Euclidean distance, see (9) as well as measuring the lateral distance between n win and all its neighboring neurons n excited using the Gaussian function, see (10). ...
Understanding, modeling, and predicting student performance in higher education poses significant challenges concerning the design of accurate and robust diagnostic models. While numerous studies attempted to develop intelligent classifiers for anticipating student achievement, they overlooked the importance of identifying the key factors that lead to the achieved performance. Such identification is essential to empower program leaders to recognize the strengths and weaknesses of their academic programs, and thereby take the necessary corrective interventions to ameliorate student achievements. To this end, our paper contributes, firstly, a hybrid regression model that optimizes the prediction accuracy of student academic performance, measured as future grades in different courses, and, secondly, an optimized multi-label classifier that predicts the qualitative values for the influence of various factors associated with the obtained student performance. The prediction of student performance is produced by combining three dynamically weighted techniques, namely collaborative filtering, fuzzy set rules, and Lasso linear regression. However, the multi-label prediction of the influential factors is generated using an optimized self-organizing map. We empirically investigate and demonstrate the effectiveness of our entire approach on seven publicly available and varying datasets. The experimental results show considerable improvements compared to single baseline models (e.g. linear regression, matrix factorization), demonstrating the practicality of the proposed approach in pinpointing multiple factors impacting student performance. As future works, this research emphasizes the need to predict the student attainment of learning outcomes.
... To confirm that the DRFs selected by the t test, regression model and the linear mixed-effect model are appropriate and to ensure these features are not highly correlated to each other, a self-organizing neural network 36 was built using the Matlab built-in neural clustering app to cluster the data based on the similarity while considering clustering in multiple dimensions. The training was performed using a batch algorithm, the slope of the DRFs over time for each data set was presented to the network before any weight was updated. ...
Changes of radiomic features over time in longitudinal images, delta radiomics, can potentially be used as a biomarker to predict treatment response. This study aims to develop a delta-radiomic process based on machine learning by (1) acquiring and registering longitudinal images, (2) segmenting and populating regions of interest (ROIs), (3) extracting radiomic features and calculating their changes (delta-radiomic features, DRFs), (4) reducing feature space and determining candidate DRFs showing treatment-induced changes, and (5) creating outcome prediction models using machine learning. This process was demonstrated by retrospectively analyzing daily non-contrast CTs acquired during routine CT-guided-chemoradiation therapy for 90 pancreatic cancer patients. A total of 2520 CT sets (28-daily-fractions-per-patient) along with their pathological response were analyzed. Over 1300 radiomic features were extracted from the segmented ROIs. Highly correlated DRFs were ruled out using Spearman correlations. Correlation between the selected DRFs and pathological response was established using linear-regression-models. T test and linear-mixed-effects-models were used to determine which DRFs changed significantly compared with first fraction. A Bayesian-regularization-neural-network was used to build a response prediction model. The model was trained using 50 patients and leave-one-out-cross-validation. Performance was judged using the area-under-ROC-curve. External independent validation was done using data from the remaining 40 patients. The results show that 13 DRFs passed the tests and demonstrated significant changes following 2–4 weeks of treatment. The best performing combination differentiating good versus bad responders (CV-AUC = 0.94) was obtained using normalized-entropy-to-standard-deviation-difference-(NESTD), kurtosis, and coarseness. With further studies using larger data sets, delta radiomics may develop into a biomarker for early prediction of treatment response.
... Diante do exposto, o presente artigo tem por objetivo, dentro do contexto de redes neurais artificiais, utilizar neurônios de uma região do Mapa de Kohonen para a classificação multirrótulo de exemplos. Para isso, será estendido o trabalho proposto por [Colombini et al. 2017], a fim de que as classes atribuídas a uma entrada de teste sejam escolhidas não apenas com base em um único neurônio vencedor, mas também com base na vizinhança ao redor desse neurônio. Utilizando conjuntos de dados artificiais e reais, esse trabalho visa investigar se a utilização de uma vizinhança de neurônios ao redor do neurônio vencedor leva a melhores resultados se comparado com a utilização de apenas um neurônio vencedor. ...
... O trabalho proposto por [Colombini et al. 2017] inspira grande parte do presente artigo. Naquele trabalho, os autores demonstraram que a tarefa de classificação multirrótulo utilizando Mapas de Kohonen é promissora, mesmo trazendo resultados de classificação utilizando apenas o neurônio vencedor. ...
... Naquele trabalho, os autores demonstraram que a tarefa de classificação multirrótulo utilizando Mapas de Kohonen é promissora, mesmo trazendo resultados de classificação utilizando apenas o neurônio vencedor. Assim, neste trabalho propomos uma extensão do método de [Colombini et al. 2017], utilizando, além do neurônio vencedor, uma vizinhança ao redor desse neurônio. · 3 ...
O problema convencional de classificação no contexto do aprendizado de máquina consiste em classificar exemplos de conjuntos de dados em categorias pré-definidas, de acordo com uma ou mais características semelhantes. Contudo, alguns conjuntos de dados possuem classes com intersecções, ou seja, exemplos podem pertencer a mais de uma classe simultaneamente. Exemplos desses problemas podem ser encontrados, por exemplo, na identificação de gêneros de livros e na classificação de imagens. Esses tipos de problemas são denominados multirrótulo. O objetivo deste artigo é propor um novo método de classificação multirrótulo com Mapas de Kohonen. A ideia é utilizar o neurônio vencedor do processo competitivo do mapa auto-organizável, juntamente com a vizinhança ao redor desse neurônio, para a classificação de dados. Assim, um novo exemplo é classificado nas classes pertencentes aos exemplos de treino mapeados para o neurônio vencedor e sua vizinhança. A linguagem Python e a biblioteca de aprendizado de máquina Scikit-Learn foram utilizadas para implementação do modelo da rede neural, para implementação das medidas de avaliação, e para a geração de conjuntos de dados sintéticos. A utilização de uma vizinhança de neurônios foi comparada com uma proposta anterior utilizando apenas um neurônio vencedor. Os resultados mostraram que a utilização de uma vizinhança ao redor do neurônio vencedor é promissora, obtendo melhores resultados.
... Most of the existing approaches of online learning tasks (for single or multi-label data) use a two-phase schema consisting of an online component, which produces summary statistics, and an offline component that uses this summary to generate clusters and predict labels for the unlabeled data (in a transductive setting). The offline component (known as batch learning) uses readily available unlabeled data to help build the predictive model, either traditionnal multi-label solutions, based on supervised learning [5] or on unsupervised learning [18], or transductive semi-supervised Multi-label approaches [5], [19]- [24]. So, batch learning can only make predictions for unlabeled data which have been used in the training phase. ...
... Our main goals are simultaneously to reduce the required amount of human effort in terms of providing ground truth label information, and to operate online in the sense that it incorporates new information directly to update and refine its internal topology. Our approach is inspired by two established graph-based methods: the Online Semi-supervised Growing Neural Gas [31], which is unable to deal with online Multilabel data, and SOM Multi-label algorithm [18], a Multi-Label version of the Unsupervised Self-Organizing Map [32]. Contrary to current neural network approaches, the proposed algorithm's topology is updated accordingly to the labels associated to each labeled data. ...
... The network is trained with all the available input patterns, with and without labels. • SOM-ML [18] (Self-Organising Map Multi-Label): a recent multi-label algorithm based on unsupervised learning through neural networks. More specifically, it explored the advantages of the self-organization ability in Self-Organizing Maps approaches, mapping input observations to neurons. ...
