A particle swarm optimization based multiobjective memetic algorithm for high-dimensional feature selection

Abstract

Feature selection, as one of the dimension reduction methods, is a crucial processing step in dealing with high-dimensional data. It tries to preserve feature subset representing the whole feature space, which aims to reduce redundancy and increase the classification accuracy. Since the two objectives are usually in conflict with each other, feature selection is modeled as a multi-objective problem. However, the high search space and discrete Pareto front makes it not easy for existing evolutionary multiobjective algorithms. Classic evolutionary computation method, which is often applied to feature selection problem straightforwardly, gradually exposes its inefficiency in searching process. Hence, a particle swarm optimization based multiobjective memetic algorithm for high-dimensional feature selection is designed in this paper to deal with above shortcomings. Its basic idea is to model feature selection as a multiobjective optimization problem by optimizing the number of features and the classification accuracy in supervised condition simultaneously, in which information entropy based initialization and adaptive local search are designed to improve the search efficiency. Moreover, a new particle velocity update rule considering both convergence and diversity of solutions is designed to update particles, and a fast discrete nondominated sorting strategy is designed to rank the Pareto solutions. These strategies enable the proposed algorithm to gain better performance on both the quality and size of feature subset. The experimental results show that the proposed algorithm can improve the quality of Pareto fronts evolved by the state-of-the-art algorithms for feature selection.