Feature Selection Using an Improved Gravitational Search Algorithm

Abstract

Feature selection is an important issue in the field of machine learning, which can reduce misleading computations and improve classification performance. Generally, feature selection can be considered as a binary optimization problem. Gravitational Search Algorithm (GSA) is a population-based heuristic algorithm inspired by Newton's laws of gravity and motion. Although GSA shows good performance in solving optimization problems, it has a shortcoming of premature convergence. In this paper, the concept of global memory is introduced and the definition of exponential Kbest is used in an improved version of GSA called IGSA. In this algorithm, the position of the optimal solution obtained so far is memorized, which can effectively prevent particles from gathering together and moving slowly. In this way, the exploitation ability of the algorithm gets improved, and a proper balance between exploration and exploitation gets established. Besides, the exponential Kbest can significantly decrease the running time. In order to solve feature selection problem, a binary IGSA (BIGSA) is further introduced. The proposed algorithm is tested on a set of standard datasets and compared with other algorithms. The experimental results confirm the high efficiency of BIGSA for feature selection.