Enhanced SparseEA for large-scale multi-objective feature selection problems

Abstract

Abstract Large-scale multi-objective feature selection problems are widely existing in the fields of text classification, image processing, and biological omics. Numerous features usually mean more correlation and redundancy between features, so effective features are usually sparse. SparseEA is an evolutionary algorithm for solving Large-scale Sparse Multi-objective Optimization Problems (i.e., most decision variables of the optimal solutions are zero). It determines feature Scores by calculating the fitness of individual features, which does not reflect the correlation between features well. In this manuscript, ReliefF was used to calculate the weights of features, with unimportant features being removed first. Then combine the weights calculated by ReliefF with Scores of SparseEA to guide the evolution process. Moreover, the Scores of features remain constant throughout all runs in SparseEA. Therefore, the fitness values of excellent and poor individuals in each iteration are used to update the Scores. In addition, difference operators of Differential Evolution are introduced into SparseEA to increase the diversity of solutions and help the algorithm jump out of the local optimal solution. Comparative experiments are performed on large-scale datasets selected from scikit-feature repository. The results show that the proposed algorithm is superior to the original SparseEA and the state-of-the-art algorithms.

Keywords

• Sparse multi-objective problems
• Large-scale feature selection
• Difference operator
• Feature weights
• SparseEA