A multi-objective feature selection based on differential evolution

This paper develops an effective multi-objective feature selection algorithm based on differential evolution (DE). In the algorithm, a randomized localization mutation based on Pareto domination is used to improve the convergence of DE. A self-adaptive crossover is proposed to dynamically assign the crossover probability of each individual. Based on it, the promising regions around good individuals will most likely be exploited. The proposed multi-objective algorithm is compared with three existing multi-objective feature selection algorithms. Experimental results show that it is a highly competitive method for solving feature selection problem.