Article
Computer Science, Information Systems
Israa Al-Badarneh, Maria Habib, Ibrahim Aljarah, Hossam Faris
Summary: This paper introduces three stochastic and metaheuristic algorithms to train MLP neural network for solving the problem of imbalanced classifications. The algorithms are evaluated using accuracy, F-score, and G-mean, and the results show that F-score and G-mean are more advantageous when the datasets are imbalanced.
JOURNAL OF KING SAUD UNIVERSITY-COMPUTER AND INFORMATION SCIENCES
(2022)
Article
Computer Science, Artificial Intelligence
Michal Koziarski
Summary: This paper proposes a unified framework for addressing the issues of oversampling and undersampling in imbalanced data, utilizing radial basis functions to preserve the original shape of class distributions and optimizing the positions of synthetic observations. Experimental results show that the proposed approach outperforms state-of-the-art resampling algorithms in handling imbalanced datasets.
PATTERN RECOGNITION
(2021)
Article
Computer Science, Artificial Intelligence
J. Hoyos-Osorio, A. Alvarez-Meza, G. Daza-Santacoloma, A. Orozco-Gutierrez, G. Castellanos-Dominguez
Summary: This paper introduces a Relevant Information-based UnderSampling (RIUS) approach to enhance classification performance for imbalanced data scenarios. Experimental results demonstrate that RIUS and its enhancement CRIUS effectively reduce information loss.
Article
Computer Science, Information Systems
Hao Tong, Changwu Huang, Leandro L. Minku, Xin Yao
Summary: This paper provides a systematic review and comprehensive empirical study of surrogate models used in single-objective SAEAs, introducing a new taxonomy and comparing the characteristics of different models through experiments. The results are helpful for researchers to select suitable surrogate models when designing SAEAs.
INFORMATION SCIENCES
(2021)
Article
Computer Science, Information Systems
Ming Zheng, Tong Li, Xiaoyao Zheng, Qingying Yu, Chuanming Chen, Ding Zhou, Changlong Lv, Weiyi Yang
Summary: A novel three-stage undersampling framework, UFFDFR, is proposed to improve classification performance on imbalanced data by removing noise and unrepresentative samples. Experiments show that UFFDFR outperformed classic and state-of-the-art clustering-based undersampling methods in terms of F-measure, G-mean, and AUC for five classification algorithms.
INFORMATION SCIENCES
(2021)
Article
Computer Science, Artificial Intelligence
Sima Mayabadi, Hamid Saadatfar
Summary: An imbalanced dataset poses challenges to learning algorithms due to the majority class dominance, and this paper proposes two density-based algorithms that use undersampling and oversampling techniques to eliminate overlap and noise, achieving balanced and normalized class distribution. These algorithms outperform other popular algorithms in various evaluation criteria.
KNOWLEDGE-BASED SYSTEMS
(2022)
Article
Computer Science, Artificial Intelligence
Eyad Elyan, Carlos Francisco Moreno-Garcia, Chrisina Jayne
Summary: Class-imbalanced datasets are common in various domains, and using class decomposition and oversampling methods can effectively reduce the dominance of majority class instances. Experimental results demonstrate the effectiveness and superiority of the proposed hybrid approach in addressing class imbalance.
NEURAL COMPUTING & APPLICATIONS
(2021)
Article
Computer Science, Artificial Intelligence
Xiaoying Xie, Huawen Liu, Shouzhen Zeng, Lingbin Lin, Wen Li
Summary: Undersampling is a widely used technique for addressing imbalanced data. Traditional methods tend to miss valuable information, leading to the development of strategies like clustering. This paper proposes a novel undersampling method based on density peaks to progressively extract instances from majority classes, using two factors to measure importance and automatically determine the optimal undersampling size. Experimental results show its superior performance compared to existing methods.
KNOWLEDGE-BASED SYSTEMS
(2021)
Article
Chemistry, Multidisciplinary
Mohamed S. Kraiem, Fernando Sanchez-Hernandez, Maria N. Moreno-Garcia
Summary: This study analyzed various factors in imbalanced data classification and derived models for automatic selection of the best resampling strategy based on dataset characteristics, covering a wide range of data conditions and performance metrics.
APPLIED SCIENCES-BASEL
(2021)
Article
Automation & Control Systems
Kaixiang Yang, Zhiwen Yu, C. L. Philip Chen, Wenming Cao, Hau-San Wong, Jane You, Guoqiang Han
Summary: The research team proposed a hybrid classifier ensemble framework that includes data transformation and an adaptive undersampling process, which effectively solves the classification problem of imbalanced data. They also designed a progressive ensemble framework to improve the performance of the classifier ensemble.
IEEE TRANSACTIONS ON SYSTEMS MAN CYBERNETICS-SYSTEMS
(2022)
Article
Computer Science, Information Systems
Zhi Chen, Jiang Duan, Li Kang, Guoping Qiu
Summary: A hybrid data-level ensemble method was developed to address the performance degradation issue caused by highly imbalanced class distribution. By integrating undersampling and oversampling, the method aims to balance data distribution and optimize the fundamental properties of the ensemble. Experimental results on 42 highly imbalanced datasets demonstrated the significant performance advantages of the proposed HD-Ensemble over other ensemble solutions.
INFORMATION SCIENCES
(2021)
Article
Green & Sustainable Science & Technology
Claudia C. Tusell-Rey, Oscar Camacho-Nieto, Cornelio Yanez-Marquez, Yenny Villuendas-Rey
Summary: This paper introduces a novel undersampling procedure for dealing with multiclass hybrid data and explores its impact on the performance of the customized naive associative classifier (CNAC). The experiments and statistical analysis demonstrate that the proposed method surpasses existing classifiers and can handle multiclass, hybrid, and incomplete data with low computational cost. Furthermore, the experiments show that CNAC benefits from data sampling, thus recommending the undersampling procedure for balancing data for CNAC.
Article
Computer Science, Information Systems
Junhai Zhai, Jiaxing Qi, Chu Shen
Summary: In this study, two binary imbalanced data classification methods, BIDC1 and BIDC2, based on diversity oversampling by generative models, are proposed to address the issue of imbalanced data classification. Experimental results demonstrate that these two methods outperform 14 state-of-the-art methods on 26 datasets.
INFORMATION SCIENCES
(2022)
Article
Chemistry, Analytical
Leehter Yao, Tung-Bin Lin
Summary: This paper introduces an effective oversampling method called evolutionary Mahalanobis distance oversampling (EMDO) for multi-class imbalanced data classification. EMDO uses a set of ellipsoids to approximate decision regions of the minority class, and integrates multi-objective particle swarm optimization (MOPSO) with the Gustafson-Kessel algorithm to generate synthetic minority samples. Results of computer simulations show that EMDO outperforms most widely used oversampling schemes.
Article
Computer Science, Artificial Intelligence
Zhan Ao Huang, Yongsheng Sang, Yanan Sun, Jiancheng Lv
Summary: Most real-life data suffer from imbalance problems, causing negative class preference behavior in neural networks. To address this issue, a new paradigm is proposed, which includes an informative undersampling strategy to solve the problem of gradient inundation and a boundary expansion strategy to alleviate the problem of insufficient empirical representation of positive samples. Experimental results show that the proposed paradigm outperforms existing methods in terms of AUC on multiple imbalanced datasets.
IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS
(2023)
Article
Computer Science, Theory & Methods
H. Bustince, R. Mesiar, J. Fernandez, M. Galar, D. Paternain, A. Altalhi, G. P. Dimuro, B. Bedregal, Z. Takac
Summary: The paper introduces a new class of functions called d-Choquet integrals, which are a generalization of the standard Choquet integral by replacing the difference in the definition with a dissimilarity function. Some d-Choquet integrals are aggregation functions, while others are not, and the conditions for this are explored in the study of their properties.
FUZZY SETS AND SYSTEMS
(2021)
Article
Computer Science, Artificial Intelligence
Francesco Zola, Lander Segurola-Gil, Jan L. Bruse, Mikel Galar, Raul Orduna-Urrutia
Summary: This work proposes a new method to address the class imbalance problem in Bitcoin entity classification by applying generative adversarial networks (GANs). By generating synthetic data to tackle the imbalance issue, GANs prove to be effective in improving accuracy and performance compared to other data preprocessing techniques.
APPLIED INTELLIGENCE
(2022)
Article
Computer Science, Information Systems
F. Zola, L. Segurola-Gil, J. L. Bruse, M. Galar, R. Orduna-Urrutia
Summary: Network traffic analysis plays a crucial role in cybersecurity by identifying anomalous and potentially dangerous connections. This study proposes a threefold approach, involving temporal dissection, data-level preprocessing, and node behavior classification, to address the challenges of analyzing temporal network traffic data. Experimental results demonstrate that the proposed method effectively reduces class imbalance and improves the performance of supervised node behavior classification, outperforming traditional anomaly detection techniques.
COMPUTERS & SECURITY
(2022)
Article
Computer Science, Artificial Intelligence
Hoang Lam Le, Ferrante Neri, Isaac Triguero
Summary: This paper investigates the optimization of instance reduction, a key stage in data mining, and proposes a Memetic Computing approach called SPMS-ALS. By integrating an Accelerated Local Search within a single-point memetic framework, SPMS-ALS achieves excellent performance while reducing runtime by up to approximately 85%, compared to other algorithms performing the same number of function calls.
SWARM AND EVOLUTIONARY COMPUTATION
(2022)
Article
Computer Science, Theory & Methods
Zdenko Takac, Mikel Uriz, Mikel Galar, Daniel Paternain, Humberto Bustince
Summary: In this work, we introduce the concept of d(G)-Choquet integral, which extends the discrete Choquet integral by incorporating a dissimilarity function to represent input differences and replacing the sum with more general functions. We demonstrate that the discrete Choquet integral and the d-Choquet integral are specific cases of the d(G)-Choquet integral. We also define interval-valued fuzzy measures and show their application in defining a monotonic interval-valued discrete Choquet integral using d(G)-Choquet integrals. The validity of this interval-valued Choquet integral is studied through an illustrative example in a classification problem.
FUZZY SETS AND SYSTEMS
(2022)
Article
Computer Science, Information Systems
Mikel Uriz, Daniel Paternain, Humberto Bustince, Mikel Galar
Summary: Fuzzy measure-based aggregations consider interactions among input source coalitions, but defining the fuzzy measure is a challenge. This paper proposes a new algorithm for learning fuzzy measure that can optimize any cost function, using advancements from deep learning frameworks. Experimental study with 58 datasets shows the effectiveness of the proposed method in optimizing cross-entropy cost for binary and multi-class classification problems, compared to other state-of-the-art methods for fuzzy measure learning.
INFORMATION SCIENCES
(2023)
Article
Physics, Applied
Sonia Elizondo, Inigo Ezcurdia, Jaime Goni, Mikel Galar, Asier Marzo
Summary: Ultrasonic fields have various functions and limitations in creating dynamic amplitude patterns. This study demonstrates how the average of multiple time-multiplexed amplitude fields improves pattern resolution and optimizes the nonlinear problem of decomposing a target amplitude field. The technique has the potential to enhance the quality of existing setups without modifying the equipment, benefiting bio-printing, haptic devices, and ultrasonic medical treatments.
APPLIED PHYSICS LETTERS
(2023)
Proceedings Paper
Computer Science, Artificial Intelligence
Iris Dominguez-Catena, Daniel Paternain, Mikel Galar
Summary: Facial Expression Recognition (FER) uses images of faces to identify the emotional state of users, allowing for a closer interaction between humans and autonomous systems. Machine learning-based models have become popular in FER but are prone to demographic bias issues. This study demonstrates the impact of gender bias in FER datasets and highlights the need for a thorough bias analysis, as global demographic balance can hide other harmful biases.
MACHINE LEARNING AND PRINCIPLES AND PRACTICE OF KNOWLEDGE DISCOVERY IN DATABASES, ECML PKDD 2022, PT I
(2023)
Proceedings Paper
Computer Science, Software Engineering
Francesco Zola, Jose Alvaro Fernandez-Carrasco, Jan Lukas Bruse, Mikel Galar, Zeno Geradts
Summary: This study proposes a novel approach that utilizes long-range images for implementing an iris verification system and uses Graph Siamese Neural Networks to predict whether they belong to the same person. The research not only describes the methodology but also evaluates the application of spectral components in improving graph extraction and classification tasks.
PROCEEDINGS OF 2022 THE 3RD EUROPEAN SYMPOSIUM ON SOFTWARE ENGINEERING, ESSE 2022
(2022)
Proceedings Paper
Computer Science, Artificial Intelligence
Hoang Lam Le, Ferrante Neri, Dario Landa-Silva, Isaac Triguero
Summary: This paper investigates a fast optimization approach for instance reduction in data science, considering both instance selection and instance generation stages. The proposed method, named APS-VSS, uses a variable solution size, accelerated objective function computation, and a single-point memetic structure for instance generation. The experiment results show that APS-VSS outperforms existing algorithms and is competitive in terms of accuracy and reduction rates, while significantly reducing the runtime.
PROCEEDINGS OF THE 2022 GENETIC AND EVOLUTIONARY COMPUTATION CONFERENCE COMPANION, GECCO 2022
(2022)
Proceedings Paper
Geosciences, Multidisciplinary
C. Ayala, C. Aranda, M. Galar
Summary: Building footprint maps are important but difficult to maintain. This study proposes a novel deep learning architecture to accurately extract building footprints from high resolution satellite imagery, bridging the gap between satellite and aerial semantic segmentation.
2022 IEEE INTERNATIONAL GEOSCIENCE AND REMOTE SENSING SYMPOSIUM (IGARSS 2022)
(2022)
Proceedings Paper
Geography, Physical
C. Ayala, C. Aranda, M. Galar
Summary: Semantic segmentation of remote sensing images is important in various practical applications, but deep learning models require a large amount of labeled data to handle unseen scenarios. This paper proposes a novel realistic multi-temporal color data augmentation technique and evaluates it in building and road semantic segmentation tasks.
XXIV ISPRS CONGRESS: IMAGING TODAY, FORESEEING TOMORROW, COMMISSION III
(2022)
Proceedings Paper
Computer Science, Artificial Intelligence
Luis Iniguez, Mikel Galar
Summary: The advancements in Big Data, Internet of Things, and Artificial Intelligence are driving the industrial revolution known as Industry 4.0. However, implementing Industry 4.0 in automated factories comes with challenges such as lack of infrastructure, financial limitations, coordination problems, and a low understanding of its implications. Many implementations focus on specific problems, leading to continuous restructuring and increased costs. To make Industry 4.0 affordable for Small and Medium-sized Enterprises (SMEs), it is necessary to create flexible and scalable Big Data architectures that take these difficulties into account.
16TH INTERNATIONAL CONFERENCE ON SOFT COMPUTING MODELS IN INDUSTRIAL AND ENVIRONMENTAL APPLICATIONS (SOCO 2021)
(2022)
Article
Computer Science, Artificial Intelligence
Jin Zhang, Zekang Bian, Shitong Wang
Summary: This study proposes a novel style linear k-nearest neighbor method to extract stylistic features using matrix expressions and improve the generalizability of the predictor through style membership vectors.
APPLIED SOFT COMPUTING
(2024)
Article
Computer Science, Artificial Intelligence
Qifeng Wan, Xuanhua Xu, Jing Han
Summary: In this study, we propose an innovative approach for dimensionality reduction in large-scale group decision-making scenarios that targets linguistic preferences. The method combines TF-IDF feature similarity and information loss entropy to address challenges in decision-making with a large number of decision makers.
APPLIED SOFT COMPUTING
(2024)
Article
Computer Science, Artificial Intelligence
Hegui Zhu, Yuchen Ren, Chong Liu, Xiaoyan Sui, Libo Zhang
Summary: This paper proposes an adversarial attack method based on frequency information, which optimizes the imperceptibility and transferability of adversarial examples in white-box and black-box scenarios respectively. Experimental results validate the superiority of the proposed method and its application in real-world online model evaluation reveals their vulnerability.
APPLIED SOFT COMPUTING
(2024)
Article
Computer Science, Artificial Intelligence
Jing Tang, Xinwang Liu, Weizhong Wang
Summary: This paper proposes a hybrid generalized TODIM approach in the Fine-Kinney framework to evaluate occupational health and safety hazards. The approach integrates CRP, dynamic SIN, and PLTSs to handle opinion interactions and incomplete opinions among decision makers. The efficiency and rationality of the proposed approach are demonstrated through a numerical example, comparison, and sensitivity studies.
APPLIED SOFT COMPUTING
(2024)
Article
Computer Science, Artificial Intelligence
Shigen Shen, Chenpeng Cai, Zhenwei Li, Yizhou Shen, Guowen Wu, Shui Yu
Summary: To address the damage caused by zero-day attacks on SIoT systems, researchers propose a heuristic learning intrusion detection system named DQN-HIDS. By integrating Deep Q-Networks (DQN) into the system, DQN-HIDS gradually improves its ability to identify malicious traffic and reduces resource workloads. Experiments demonstrate the superior performance of DQN-HIDS in terms of workload, delayed sample queue, rewards, and classifier accuracy.
APPLIED SOFT COMPUTING
(2024)
Article
Computer Science, Artificial Intelligence
Song Deng, Qianliang Li, Renjie Dai, Siming Wei, Di Wu, Yi He, Xindong Wu
Summary: In this paper, we propose a Chinese text classification algorithm based on deep active learning for the power system, which addresses the challenge of specialized text classification. By applying a hierarchical confidence strategy, our model achieves higher classification accuracy with fewer labeled training data.
APPLIED SOFT COMPUTING
(2024)
Article
Computer Science, Artificial Intelligence
Kaan Deveci, Onder Guler
Summary: This study proves the lack of robustness in nonlinear IF distance functions for ranking intuitionistic fuzzy sets (IFS) and proposes an alternative ranking method based on hypervolume metric. Additionally, the suggested method is extended as a new multi-criteria decision making method called HEART, which is applied to evaluate Turkey's energy alternatives.
APPLIED SOFT COMPUTING
(2024)
Article
Computer Science, Artificial Intelligence
Fu-Wing Yu, Wai-Tung Ho, Chak-Fung Jeff Wong
Summary: This research aims to enhance the energy management in commercial building air-conditioning systems, specifically focusing on chillers. Ridge regression is found to outperform lasso and elastic net regression when optimized with the appropriate hyperparameter, making it the most suitable method for modeling the system coefficient of performance (SCOP). The key variables that strongly influence SCOP include part load ratios, the operating numbers of chillers and pumps, and the temperatures of chilled water and condenser water. Additionally, July is identified as the month with the highest potential for performance improvement. This study introduces a novel approach that balances feature selection, model accuracy, and optimal tuning of hyperparameters, highlighting the significance of a generic and simplified chiller system model in evaluating energy management opportunities for sustainable operation. The findings from this research can guide future efforts towards more energy-efficient and sustainable operations in commercial buildings.
APPLIED SOFT COMPUTING
(2024)
Article
Computer Science, Artificial Intelligence
Xiaoyan Chen, Yilin Sun, Qiuju Zhang, Xuesong Dai, Shen Tian, Yongxin Guo
Summary: In this study, a method for dynamically non-destructive grasping of thin-skinned fruits is proposed. It utilizes a multi-modal depth fusion convolutional neural network for image processing and segmentation, and combines the evaluation mechanism of optimal grasping stability and the forward-looking non-destructive grasp control algorithm. The proposed method greatly improves the comprehensive performance of grasping delicate fruits using flexible hands.
APPLIED SOFT COMPUTING
(2024)
Article
Computer Science, Artificial Intelligence
Yuxuan Yang, Siyuan Zhou, He Weng, Dongjing Wang, Xin Zhang, Dongjin Yu, Shuiguang Deng
Summary: The study proposes a novel model, POIGDE, which addresses the challenges of data sparsity and elusive motives by solving graph differential equations to capture continuous variation of users' interests. The model learns interest transference dynamics using a time-serial graph and an interval-aware attention mechanism, and applies Siamese learning to directly learn from label representations for predicting future POI visits. The model outperforms state-of-the-art models on real-world datasets, showing potential in the POI recommendation domain.
APPLIED SOFT COMPUTING
(2024)
Article
Computer Science, Artificial Intelligence
S. Karthika, P. Rathika
Summary: The widespread development of monitoring devices in the power system has generated a large amount of power consumption data. Storing and transmitting this data has become a significant challenge. This paper proposes an adaptive data compression algorithm based on the discrete wavelet transform (DWT) for power system applications. It utilizes multi-objective particle swarm optimization (MO-PSO) to select the optimal threshold. The algorithm has been tested and outperforms other existing algorithms.
APPLIED SOFT COMPUTING
(2024)
Article
Computer Science, Artificial Intelligence
Jiaqi Guo, Haiyan Wu, Xiaolei Chen, Weiguo Lin
Summary: In this study, an adaptive SV-Borderline SMOTE-SVM algorithm is proposed to address the challenge of imbalanced data classification. The algorithm maps the data into kernel space using SVM and identifies support vectors, then generates new samples based on the neighbors of these support vectors. Extensive experiments show that this method is more effective than other approaches in imbalanced data classification.
APPLIED SOFT COMPUTING
(2024)
Article
Computer Science, Artificial Intelligence
Qiumei Zheng, Linkang Xu, Fenghua Wang, Yongqi Xu, Chao Lin, Guoqiang Zhang
Summary: This paper proposes a new semantic segmentation network model called HilbertSCNet, which combines the Hilbert curve traversal and the dual pathway idea to design a new spatial computation module to address the problem of loss of information for small targets in high-resolution images. The experiments show that the proposed network performs well in the segmentation of small targets in high-resolution maps such as drone aerial photography.
APPLIED SOFT COMPUTING
(2024)
Article
Computer Science, Artificial Intelligence
Mojtaba Ashour, Amir Mahdiyar
Summary: Analytic Hierarchy Process (AHP) is a widely applied technique in multi-criteria decision-making problems, but the sheer number of AHP methods presents challenges for scholars and practitioners in selecting the most suitable method. This paper reviews articles published between 2010 and 2023 proposing hybrid, improved, or modified AHP methods, classifies them based on their contributions, and provides a comprehensive summary table and roadmap to guide the method selection process.
APPLIED SOFT COMPUTING
(2024)
Review
Computer Science, Artificial Intelligence
Gerardo Humberto Valencia-Rivera, Maria Torcoroma Benavides-Robles, Alonso Vela Morales, Ivan Amaya, Jorge M. Cruz-Duarte, Jose Carlos Ortiz-Bayliss, Juan Gabriel Avina-Cervantes
Summary: Electric power system applications are complex optimization problems. Most literature reviews focus on studying electrical paradigms using different optimization techniques, but there is a lack of review on Metaheuristics (MHs) in these applications. Our work provides an overview of the paradigms underlying such applications and analyzes the most commonly used MHs and their search operators. We also discover a strong synergy between the Renewable Energies paradigm and other paradigms, and a significant interest in Load-Forecasting optimization problems. Based on our findings, we provide helpful recommendations for current challenges and potential research paths to support further development in this field.
APPLIED SOFT COMPUTING
(2024)