Article
Computer Science, Artificial Intelligence
Binbin Pan, Wen-Sheng Chen, Liping Deng, Chen Xu, Xiaobo Zhou
Summary: The selection of proper classifiers for a given data set is challenging, and the critical problem is how to extract features. This paper proposes a new method that preserves the geometrical structure and characterizes the decision boundary of a data set. The extracted features can recover the same Euclidean geometrical structure as the original data set. An efficient algorithm is presented to compute the similarity between data set features, and the impact of feature similarity on the performance of the support vector machine is theoretically analyzed. Empirical results demonstrate the effectiveness of the proposed method in finding suitable classifiers.
NEURAL COMPUTING & APPLICATIONS
(2023)
Article
Computer Science, Artificial Intelligence
Khalid Benabdeslem, Dou El Kefel Mansouri, Raywat Makkhongkaew
Summary: This paper focuses on the co-selection of instances and features in the semi-supervised learning scenario. It proposes a unified framework, called sCOs, that integrates labeled and unlabeled parts into the co-selection process. Two efficient algorithms are proposed and experimental results validate the effectiveness of the method.
IEEE TRANSACTIONS ON KNOWLEDGE AND DATA ENGINEERING
(2022)
Article
Computer Science, Information Systems
Mingzhao Wang, Henry Han, Zhao Huang, Juanying Xie
Summary: It is proposed in this paper to detect the informative features for high dimensional data with a small number of samples through two unsupervised spectral feature selection algorithms. These algorithms group features using an advanced Self-Tuning spectral clustering algorithm and detect the global optimal feature clusters through feature ranking techniques. Extensive experiments demonstrate the effectiveness of the proposed algorithms, especially the one based on cosine similarity feature ranking technique. The detected features have strong discriminative capabilities, making them suitable for building reliable and explainable AI systems, particularly in medical diagnostic systems.
FRONTIERS OF COMPUTER SCIENCE
(2023)
Article
Computer Science, Artificial Intelligence
Abdul Wahid, Dost Muhammad Khan, Ijaz Hussain, Sajjad Ahmad Khan, Zardad Khan
Summary: A novel robust unsupervised feature selection method, UFS-RDR, is proposed to improve feature selection performance by minimizing the graph regularized weighted data reconstruction error function, using Mahalanobis distance to detect outliers and determine Huber-type weight function. The experimental results show that UFS-RDR outperforms non-robust methods in the presence of contamination in unlabeled data.
EXPERT SYSTEMS WITH APPLICATIONS
(2022)
Article
Biochemical Research Methods
Fahimeh Motamedi, Horacio Perez-Sanchez, Alireza Mehridehnavi, Afshin Fassihi, Fahimeh Ghasemi
Summary: This article discusses two approaches for quantitative structure-activity prediction studies, focusing on identifying appropriate molecular descriptors and predicting the biological activities of designed compounds. The use of LASSO-random forest algorithm is shown to significantly improve output correlation, reduce implementation time and model complexity, while maintaining prediction accuracy.
Article
Genetics & Heredity
Juanying Xie, Mingzhao Wang, Shengquan Xu, Zhao Huang, Philip W. Grant
Summary: In this paper, an unsupervised feature selection technique called SCFS is proposed to address challenges in genomic data analysis caused by high dimensionality and imbalanced class distribution. By defining discernibility and independence of features, an optimal feature subset with high classification capability is identified for KNN and SVM classifiers, leading to improved results in genomic datasets analysis.
FRONTIERS IN GENETICS
(2021)
Article
Automation & Control Systems
Shuping Zhao, Jigang Wu, Bob Zhang, Lunke Fei, Shuyi Li, Pengyang Zhao
Summary: This article proposes a novel adaptive graph embedded preserving projection learning method. By combining sparse graph learning and projection learning, it achieves feature extraction and selection, and has been proven effective and competitive through experiments.
IEEE TRANSACTIONS ON SYSTEMS MAN CYBERNETICS-SYSTEMS
(2023)
Article
Computer Science, Information Systems
Wenbin Qian, Yinsong Xiong, Jun Yang, Wenhao Shu
Summary: Feature selection is crucial in machine learning and data mining, and traditional methods may not be suitable for label distribution learning. This paper proposes a novel feature selection algorithm for label distribution learning, which utilizes neighborhood granularity, correlation coefficient, and sparse learning to improve effectiveness. Experimental results show that the proposed method outperforms five state-of-art algorithms on twelve datasets.
INFORMATION SCIENCES
(2022)
Article
Chemistry, Multidisciplinary
Jun Cai, Linge Fan, Xin Xu, Xinrong Wu
Summary: This paper proposes unsupervised and supervised feature selection methods for incomplete data using L-2, L-1 norm and reconstruction error minimization methods. Experimental studies demonstrate the effectiveness of the proposed methods.
APPLIED SCIENCES-BASEL
(2022)
Article
Multidisciplinary Sciences
Xingjian Chen, Zifan Zhu, Weitong Zhang, Yuchen Wang, Fuzhou Wang, Jianyi Yang, Ka-Chun Wong
Summary: Predicting human diseases from microbiome data is important in medical applications. Existing methods often overlook the abundance profiles of known and unknown microbial organisms, as well as the taxonomic relationships among them, resulting in information loss. To address these issues, we developed a comprehensive machine learning framework called MetaDR that combines deep learning and various information sources to predict human diseases.
Article
Environmental Sciences
Qi Liu, Shibiao Xu, Jun Xiao, Ying Wang
Summary: This paper introduces a novel sharp-feature-preserving reconstruction framework based on primitive detection, which accurately segments primitive patches, fits meshes in each patch, and splits overlapping meshes at the triangle level to ensure true sharpness and obtain lightweight mesh models. Experimental results show that our framework outperforms both the state-of-the-art learning-based primitive detection methods and traditional reconstruction methods. Moreover, our designed modules are plug-and-play, and can be combined with other point cloud processing tasks to achieve high-fidelity results.
Article
Automation & Control Systems
Hui Hou, Hongquan Ji
Summary: A novel feature selection strategy is proposed to improve the multiclass support vector data description (SVDD) algorithm for planetary gearbox fault diagnosis. By selecting features sensitive to faults and developing an improved multiclass SVDD algorithm, the fault diagnosis task is effectively completed.
CONTROL ENGINEERING PRACTICE
(2021)
Article
Engineering, Electrical & Electronic
Zihao Song, Peng Song
Summary: Feature selection is a fundamental and challenging topic in machine learning and pattern recognition, and unsupervised feature selection methods have received extensive attention. In this article, a novel latent energy preserving embedding method is proposed for unsupervised feature selection, which utilizes self-representation learning strategy and graph Laplacian for mining manifold information and selects features using l(2,1)-norm. Extensive experiments on real-world datasets validate the effectiveness of the proposed method.
DIGITAL SIGNAL PROCESSING
(2022)
Article
Computer Science, Artificial Intelligence
Mohammad Amin Pirbonyeh, Mohammad Amin Shayegan, Gholamreza Sotudeh, Shahab Shamshirband
Summary: Transfer Learning (TL) algorithms are effective methods for improving classifier learning by utilizing source domain knowledge in the target domain. Reducing the difference in feature space and distribution between domains is crucial for enhancing TL algorithms. Existing methods often employ complex computational structures but overlook the preservation of data topology. This paper proposes a unified framework called FN-DTP, which addresses heterogeneous domain adaptation problems by combining feature normalization, distribution reduction, and topology preservation, resulting in improved TL algorithm performance.
KNOWLEDGE-BASED SYSTEMS
(2022)
Article
Computer Science, Artificial Intelligence
Yuang Shi, Chen Zu, Mei Hong, Luping Zhou, Lei Wang, Xi Wu, Jiliu Zhou, Daoqiang Zhang, Yan Wang
Summary: Multimodal classification methods using different modalities have advantages over traditional single-modality-based ones for the diagnosis of Alzheimer's disease and mild cognitive impairment. This paper proposes a novel multimodal feature selection method called ASMFS, which performs adaptive similarity learning and feature selection simultaneously, and demonstrates its effectiveness and superiority over other state-of-the-art approaches for multi-modality classification of AD/MCI.
PATTERN RECOGNITION
(2022)