Fuzzy C-means clustering-based multi-label feature selection via weighted neighborhood mutual information

As an effective soft computing model, fuzzy C-means (FCM) clustering can efficiently deal with uncertainty problems; however, it is usually sensitive to the selection of the initial cluster center and may fall into local optimization for classification; and existing multi-label feature selection algorithms ignore the potential relationship between labels and samples. To solve these problems, this article develops hybrid soft computing methodologies for feature selection employing FCM clustering and weighted neighborhood-based mutual information on multi-label classification tasks. First, cluster centroids are initialized stochastically via the whale optimization algorithm (WOA), and then the cluster center is updated in conjunction with the FCM clustering during the iterative process. By fusing soft computing methods of WOA and FCM clustering, this combination converges quickly and effectively avoids falling into the local optimal solution. Secondly, based on the membership degrees of each sample derived from FCM clustering, an association matrix is established, followed by a fuzzy synthesis operation to formulate a label enhancement strategy. Thirdly, with the advantage of handling unbalanced labels via neighborhood mutual information, feature weight is applied to calculate the weighted neighborhood of samples by calculating maximum information coefficient, and the redundancy among candidate and selected features is quantified through the application of weighted neighborhood-based mutual information. Finally, a feature selection algorithm utilizing label enhancement and weighted neighborhood-based mutual information is developed, and those competitive experiments will be carried out across 11 multi-label datasets. These results demonstrate that this methodology significantly enhances the classification efficacy on multi-label datasets.