A hybrid particle swarm optimization for feature subset selection by integrating a novel local search strategy.

The proposed method uses a local search technique which is embedded in particle swarm optimization (PSO) to select the reduced sized and salient feature subset. The goal of the local search technique is to guide the PSO search process to select distinct features by using their correlation information. Therefore, the proposed method selects the subset of features with reduced redundancy. A hybrid feature selection method based on particle swarm optimization is proposed.Our method uses a novel local search to enhance the search process near global optima.The method efficiently finds the discriminative features with reduced correlations.The size of final feature set is determined using a subset size detection scheme.Our method is compared with well-known and state-of-the-art feature selection methods. Feature selection has been widely used in data mining and machine learning tasks to make a model with a small number of features which improves the classifier's accuracy. In this paper, a novel hybrid feature selection algorithm based on particle swarm optimization is proposed. The proposed method called HPSO-LS uses a local search strategy which is embedded in the particle swarm optimization to select the less correlated and salient feature subset. The goal of the local search technique is to guide the search process of the particle swarm optimization to select distinct features by considering their correlation information. Moreover, the proposed method utilizes a subset size determination scheme to select a subset of features with reduced size. The performance of the proposed method has been evaluated on 13 benchmark classification problems and compared with five state-of-the-art feature selection methods. Moreover, HPSO-LS has been compared with four well-known filter-based methods including information gain, term variance, fisher score and mRMR and five well-known wrapper-based methods including genetic algorithm, particle swarm optimization, simulated annealing and ant colony optimization. The results demonstrated that the proposed method improves the classification accuracy compared with those of the filter based and wrapper-based feature selection methods. Furthermore, several performed statistical tests show that the proposed method's superiority over the other methods is statistically significant.