An improved electromagnetism-like method for feature selection

Abstract

The paper presents an improved electromagnetism-like method (EML) for solving the feature selection problem. The objective function of EML is calculated as a classification accuracy, while the feature reduction rate is used as a comparison index. Efficient local search procedure and a solution scaling mechanism enforce better search process exploitation. The speedup of the objective function evaluation is achieved by introducing caching procedure for EML. EML is tested in two separate experiments, the first one is based on 13, and the second on 6 real life instances. The results show that the presented approach outperforms previously introduced EML algorithm and genetic algorithm in 10 out of 13 cases, with respect to the feature reduction rate. The running times are in some cases up to two orders of magnitude shorter. In the second experiment, EML performed better than two variants of particle swarm optimization technique in 3 out of 6 cases.