Yazar "Ju, Hengrong" seçeneğine göre listele

Listeleniyor 1 - 2 / 2
  • Küçük Resim Yok
    Öğe
    Dual-Channel Fuzzy Interaction Information Fused Feature Selection With Fuzzy Sparse and Shared Granularities
    (IEEE-INST ELECTRICAL ELECTRONICS ENGINEERS INC, 14.11.2024) Ju, Hengrong; Fan, Xiaoxue; Ding, Weiping; Huang, Jiashuang; Xu, Suping; Yang, Xibei; Pedrycz, Witold
    Fuzzy information granularity is an effective granular computation approach for feature evaluation and selection. However, most existing methods rely on a single granulation channel, neglecting different granularity representations. In this article, a novel dual-channel fuzzy interaction information fused feature selection with fuzzy sparse and shared granularities is proposed. It mainly comprises the following three parts. First, a dual-channel framework is introduced to construct the fuzzy information granularity from two different strategies. One channel employs sparse mutual strategy to form the sparse representation-based fuzzy information granularity, while the other constructs the fuzzy shared information granularity with a novel fuzzy semi-ball. Second, in each channel, the criteria of maximum relevancy, minimum redundancy, and maximum interaction is adopted to access feature correlation and perform feature ranking. Third, the two feature sequences derived from the dual-channel are fused to form a final feature sequence based on the within-class and between-class mechanism. To validate the efficacy of the proposed method, experimental validations on 15 datasets and schizophrenia data are conducted. The results show that the proposed method outperforms other algorithms in classification accuracy and statistical analysis. Moreover, its superiority regarding accuracy can be demonstrated in the experiments of schizophrenia detection, where it performs well in recognizing schizophrenia through visual interpretation.
  • Küçük Resim Yok
    Öğe
    Multi-association evidential feature selection and its application to identifying schizophrenia
    (Elsevier Inc., 2024) Ju, Hengrong; Fan, Xiaoxue; Ding, Weiping; Huang, Jiashuang; Pedrycz, Witold; Yang, Xibei
    Granular Computing (GrC)-based feature selection can remove redundant features from a massive amount of data and improve the efficiency of information processing. However, the existing method of neighborhood-based information granule only considers the distance between samples, ignoring other significant relationships existing between them. To fill this gap, this paper proposes a novel feature selection approach based on two-step multi-association neighborhood evidence entropy. This approach is constructed in three phases. Firstly, adaptive k value corresponding to each sample in sparse representation method is determined. Sparse correlation and distance measure are fused to form a multi-association information granule. Then, the samples in the multi-association information granule are estimated and weak-related information is removed to constitute a two-step multi-association information granule. Secondly, sparse correlation information is processed using Dempster-Shafer evidence theory, and a new credibility-based function is developed. In addition, the credibility is used to construct a novel neighborhood evidence entropy, which can effectively reflect the uncertainty of data. Thirdly, the proposed neighborhood evidence entropy is applied to assess the importance of features. As a result, several vital features are selected. The experimental results on twelve datasets demonstrate that the effectiveness of the proposed method is superior to other algorithms in construction of information granules and classification accuracy, respectively. Finally, the proposed method is applied to the selection of brain regions in schizophrenia. It can effectively analyze the lesions of schizophrenia and improve the prediction of the disorder. The code is available at https://github.com/fxx-Aurora/TMAE-FS/tree/main. © 2024 Elsevier Inc.