Improved Binary Particle Swarm Optimization Based on Multi-Exemplar and Forgetting Ability Applied to Feature Selection

Abstract

Feature selection can eliminate irrelevant, redundant, and misleading features in the data set, and improve the classification performance of machine learning algorithms while reducing their computational consumption, effectively avoiding "dimensional disasters". The multi-exemplar particle swarm optimization algorithm with forgetting capability (XPSO) has achieved good performance in function optimization, but has not been applied to the feature selection problem of binary variables. In this paper, an XPSO based on binary encoding, named Binary XPSO (BXPSO) is proposed to solve the optimal feature subset. The algorithm also proposes a local search strategy to improve the forgetting ability of different particles, balancing the local exploitation and global exploration of the algorithm. To verify the effectiveness of the proposed algorithm, multiple sets of simulation experiments with different perspectives are conducted on the proposed method in this paper using classical datasets from the UCI machine learning repository constituting feature selection problems of different dimensions. The experimental results show that the algorithm has competitive advantages in terms of classification accuracy and computational performance.