Hybrid improved gravitional search algorithm and kernel based extreme learning machine method for classification problems

Abstract

In this paper, we hybridize the improved gravitational search algorithm (IGSA) with kernel based extreme learning machine (KELM) method. Based on this, a novel hybrid system IGSA-KELM is proposed to improve the generalization performance for classification problems. In this system, IGSA is designed by combining the search strategy of particle swarm optimization and GSA to effectively reduce the problem of slow convergence rate, moreover, the continuous-value IGSA and binary IGSA are integrated in one algorithm in order to optimize the KELM parameters and feature subset selection simultaneously. This proposed hybrid algorithm is evaluated on several well-known UCI machine learning datasets. The results indicate that the superiority of the proposed model in terms of classification accuracy. Our hybrid method not only can select the most relevant feature subset, but also achieves a high classification accuracy over other similar state-of-the-art classifier systems.