Fuzzt Set-Based Kernel Extreme Learning Machine Autoencoder for Multi-Label Classification

• Published in: 2021 International Conference on Machine Learning and Cybernetics (ICMLC) pp. 1 - 6
• Main Authors: Zhang, Qingshuo, Tsang, Eric C. C., Hu, Meng, He, Qiang, Chen, Degang
• Format: Conference Proceeding
• Language: English
• Published: IEEE 04-12-2021
• Subjects: Autoencoder; Classification algorithms; Correlation; Extreme learning machines; Feature extraction; Fuses; Fuzzy Set; Fuzzy set theory; Kernel Extreme Learning Machine; Machine learning; Multi-label Learning
• ISSN: 2160-1348
• DOI: 10.1109/ICMLC54886.2021.9737260

The multi-label learning algorithm based on an extreme learning machine has the advantage of high efficiency and generalization ability, but its classification ability is weak due to ignoring the correlation between features and labels. Accordingly, in this paper, the fuzzy set-based kernel extreme learning machine autoencoder for multi-label classification (KELM-AE-fuzzy) is proposed. Firstly, the correlation between features and labels is analyzed based on fuzzy set theory, and the correlation label membership matrix and label completion matrix are constructed. Then, the kernel extreme learning machine autoencoder is used to fuse the correlation label membership matrix with the original feature space and generate the reconstructed feature space. Eventually, kernel extreme learning machine (KELM) is used as a classifier, where the label matrix is used with the label completion matrix. Comparative experiments on several multi-label datasets demonstrate that KELM-AE-fuzzy outperforms other multi-label algorithms, and the effectiveness of the proposed algorithm is verified.