Abstract
Feature selection is one of the important activities in various fields such as computer vision and pattern recognition. In this paper, an improved version of the binary gravitational search algorithm (BGSA) is proposed and used as a tool to select the best subset of features with the goal of improving classification accuracy. By enhancing the transfer function, we give BGSA the ability to overcome the stagnation situation. This allows the search algorithm to explore a larger group of possibilities and avoid stagnation. To evaluate the proposed improved BGSA (IBGSA), classification of some well known datasets and improving the accuracy of CBIR systems are experienced. Results are compared with those of original BGSA, genetic algorithm (GA), binary particle swarm optimization (BPSO), and electromagnetic-like mechanism. Comparative results confirm the effectiveness of the proposed IBGSA in feature selection.