Feature Selection via Label Enhancement and Weighted Neighborhood Mutual Information for Multilabel Data

This work presents a multilabel feature selection approach via label enhancement and weighted neighborhood mutual information. First, the Fuzzy C-Means (FCM) clustering is optimized by the Whale Optimization Algorithm (WOA) to obtain the initial value of the cluster centers, and then in the iterative process, the FCM clustering algorithm is updated to ensure fast convergence and avoid local optimization. Secondly, the association matrix is constructed through the membership degree of each sample obtained by the FCM clustering, and a fuzzy synthesis operation is performed to obtain the label enhancement strategy. Thirdly, label weights are introduced into the traditional neighborhood mutual information to improve the handling effect of imbalanced labels. Feature weights are calculated via the maximum information coefficient to determine the weighted sample neighborhoods, and this weighted neighborhood mutual information assesses redundancy between the candidate and selected features. Finally, a feature selection algorithm via weighted neighborhood mutual information is designed for multilabel data classification. Those comparative experiments are performed on 11 multilabel datasets. The experimental results show that the constructed algorithm effectively improves the classification effect of multilabel datasets.