Novel self-adjusted particle swarm optimization algorithm for feature selection

Bo Wei

doi:10.1007/s00607-020-00891-w

Return

Novel self-adjusted particle swarm optimization algorithm for feature selection

2021-01-21

PRE

DOI：10.1007/s00607-020-00891-w

Original

Original request for help

Save

Abstract

En 中文

Due to the ever increasing number of features in the most practical application fields, i.e, expert and intelligent systems, feature selection (FS) has become a promising pre-processing step for a particular task (i.e., classification and regression) in the last few decades. FS aims at selecting the optimal feature subset from the original feature dataset by removing redundant and irrelevant features, which improve the performance of the learning models. In this paper, a novel self-adjusted particle swarm optimization algorithm (SAPSO) is proposed for selecting the optimal feature subset for classification datasets. In SAPSO, we make three improvements: First, a new learning model of particles, which can extract much more useful knowledge from multiple information providers, is used to enhance the diversity of particles. Second, one-flip neighborhood search strategy is adopted to strengthen the local search ability of a swarm when the swarm enters a period of stagnation. Finally, a population replacement process is conducted, which bases on the part of new particles generated by the neighborhood search strategy, to enhance the diversity of the swarm. Moreover, the k-nearest neighbor method is used as a classifier to evaluate the classification accuracy of a particle. The proposed method is benchmarked on 10 well-known UCI datasets and the results are compared with 9 state-of-the-art wrapper-based FS methods. From the results, it is observed that the proposed approach significantly outperforms others on most the 10 datasets.

Keywords:

Feature selection

Combinatorial optimization

Particle swarm optimization

Neighborhood search strategy

AI Summary

Key information extracted from the uploaded paper, including a brief overview, abstract, background, key highlights, visual analysis, and future outlook.