Fine-Grain Knowledge Transfer-based Multitask Particle Swarm Optimization with Dual Clustering-based Task Generation for High-Dimensional Feature Selection

Xin-Yu Wang,Qi-Te Yang,Yi Jiang,Kay Chen Tan,Jun Zhang,Zhi-Hui Zhan

DOI: https://doi.org/10.1145/3638529.3654023

2024-01-01

Abstract:Evolutionary multitasking (EMT), as a very popular research topic in the evolutionary computation community, has been used to solve high-dimensional FS problems and has shown good performance recently. However, most of the existing EMT-based methods still have two drawbacks. First, they only consider using filter-based task generation strategies to retain highly relevant features for generating the additional tasks, whereas the redundancy between features is ignored. Second, they always consider a complete variable vector (e.g., global optimum or mean positional information of a population at current generation) as positive knowledge and transfer it, which greatly weakens the variety of transferred knowledge and increases the possibility of falling into local optimality. To deal with these two drawbacks, we propose a new EMT-assisted multitask particle swarm optimization (MPSO) algorithm with two innovations for high-dimensional FS. First, we propose a dual clustering-based task generation strategy to generate tasks by considering both feature relevance and redundancy. Second, we propose a fine-grain knowledge transfer strategy to realize explicit transfer of knowledge between different tasks. Experimental results on 15 public datasets show the effectiveness and competitiveness of our proposed MPSO algorithm over other state-of-the-art FS methods in dealing with high-dimensional FS problems.