Evolutionary Sparse Coding and Graph Regularisation for Embedded Multi-label Feature Selection

Kaan Demir, Bach Hoai Nguyen, Bing Xue, Mengjie Zhang

Abstract: Multi-label datasets often possess hundreds of irrelevant or redundant features that can negatively affect classification performance over multiple co-occuring class labels, necessitating feature selection. Sparsity-based multi-label feature selection has garnered notable attention due to its effectiveness at identifying irrelevant and redundant features. However, most existing sparsity-based feature selection methods are gradient-based and are prone to inconsistency or premature convergence. This paper proposes novel evolutionary sparsity-based approaches for embedded multi-label feature selection. The proposed methods can achieve state-of-the-art classification performance.