Particle Swarm Optimisation and Statistical Clustering for Feature Selection

Feature selection is an important issue in classification, but it is a difficult task due to the large search space and feature interaction. Statistical clustering methods, which consider feature interaction, group features into different feature clusters. This paper investigates the use of statistical clustering information in particle swarm optimisation (PSO) for feature selection. Two PSO based feature selection algorithms are proposed to select a feature subset based on the statistical clustering information. The new algorithms are examined and compared with a greedy forward feature selection algorithm on seven benchmark datasets. The results show that the two algorithms can select a much smaller number of features and achieve similar or better classification performance than using all features. One of the new algorithms that introduces more stochasticity achieves the best results and outperforms all other methods, especially on the datasets with a relatively large number of features.