Binary Sparrow Search Algorithm for Feature Selection

Abstract

The sparrow search algorithm (SSA) is a novel intelligent optimization algorithm that simulates the foraging and anti-predation behavior of sparrows. The sparrow search algorithm (SSA) can optimize continuous problems, but in reality many problems are binary problems. In this paper, the binary sparrow search algorithm (BSSA) is proposed to solve binary optimization problems, such as feature selection. The transfer function is crucial to BSSA and it directly affects the performance of BSSA. This paper proposes three new transfer functions to improve the performance of BSSA. Mathematical analysis revealed that the original SSA scroungers position update equation is no longer adapted to BSSA. This paper improves the position update equation. We compared BSSA with BPSO, BGWO, and BBA algorithms, and tested on 23 benchmark functions. In addition, statistical analysis of the experimental results, Friedman test and Wilcoxon rank-sum test were performed to verify the effectiveness of BSSA. Finally, the algorithm was used to successfully implement feature selection and obtain satisfactory results in the UCI data set.