Dynamic Multi-Level Competition Learning-Based Dual-Task Optimization for High-Dimensional Feature Selection

Abstract

Feature selection (FS) is a critical task in data science and machine learning, presenting significant challenges in high-dimensional settings due to the complexity and noise inherent in large feature sets. To address these issues, this paper proposes a Dynamic Multi-Level Competition Learning-Based Dual-Task Optimization (DMLC-DTO) method. The approach introduces a dual-task generation strategy that uses the Fisher Score to generate sub-tasks, which aid the primary task in exploring the feature space more effectively and accelerating feature selection. The Dynamic Multi-Level Competition Learning-Based Optimization mechanism enhances population diversity by organizing particles into hierarchical levels, with lower-tier particles learning from those at higher tiers. This hierarchical structure is integrated with traditional Competitive Swarm Optimization (CSO) and with a dynamic factor regulating the balance between exploration and convergence. Furthermore, the Multi-Winner Based Knowledge Transfer method encourages inter-task learning by allowing particles at the same level across tasks to exchange knowledge and facilitate information transfer. Experiments on 13 high-dimensional, real-world datasets confirm the effectiveness and robustness of DMLC-DTO, showcasing its competitive performance in feature selection tasks.

Keywords

• Feature selection (FS)
• high-dimensional
• fisher score
• competitive swarm optimization (CSO)