A Multi-Objective Feature Selection Algorithm Based on a New Feature Subsets Adjustment Strategy

The main purpose of feature selection (FS) is to reduce the data dimension and reduce the classification error rate. Therefore, it can be regarded as a multi-objective optimization problem. Evolutionary algorithms have been successfully applied to solve feature selection problems. However, with the increase of data dimension, high-dimensional feature selection may encounter many difficulties. e.g. How to generate high-quality offspring considering both convergence and diversity. How to cope with a large number of duplicated or similar solutions in the objective space, which fall into local optimum. Therefore, this paper proposes a novel feature subsets adjustment strategy. The proposed feature subset adjustment strategy can recombine the solutions of the same number of features and expand the search range, which can avoid falling into local optima to a certain extent, and further eliminate potential redundant features and accelerate convergence. In the experiment, the proposed strategy-improved algorithm MOFS-SA is compared with the other five algorithms in nine high-dimensional data sets, and the non-dominant solution set is visualized. The results show that MOFS-SA has a good performance on the high-dimensional feature selection problem. We prove that the feature subset adjustment strategy proposed in this paper can improve the search ability of the algorithm, ensure the diversity and further accelerate the convergence rate of the algorithm.