Particle swarm optimization for feature selection with adaptive mechanism and new updating strategy

Abstract

Feature selection is an important data preprocessing technique in the emerging field of artificial intelligence and data mining which aims at finding a small set of features from the original dataset with predetermined targets. Particle swarm optimization (PSO) has been widely used to address feature selection problems because of its easy implementation, efficiency and simplicity. However, in high-dimensional problems, selecting the discriminative features with a higher correct classification rate is limited. To solve the issue above, a particle swarm optimization method with adaptive mechanism and new updating strategy is proposed to choose best features to improve the correct classification rate. The proposed approach, named as EPSO, is verified and compared with other three meta-heuristic algorithms and four recent PSO-based feature selection methods. The experimental results and statistical tests have proved the efficiency and feasibility of the EPSO approach in obtaining higher classification accuracy along with smaller number of features. Therefore, the proposed EPSO algorithm can be successfully used as a novel feature selection strategy.