Feature subset selection by gravitational search algorithm optimization

Abstract

A new method for feature subset selection in machine learning, FSS-MGSA (Feature Subset Selection by Modified Gravitational Search Algorithm), is presented. FSS-MGSA is an evolutionary, stochastic search algorithm based on the law of gravity and mass interactions, and it can be executed when domain knowledge is not available. A wrapper approach, over Naive-Bayes, ID3, K-Nearest Neighbor and Support Vector Machine learning algorithms, is used to evaluate the goodness of each visited solution. The key to the success of the MGSA is to utilize the piecewise linear chaotic map for increasing its diversity of species, and to use sequential quadratic programming for accelerating local exploitation. Promising results are achieved in a variety of tasks where domain knowledge is not available. The experimental results show that the proposed method has the ability of selecting the discriminating input features correctly and can achieve high accuracy of classification, which is comparable to or better than well-known similar classifier systems. Furthermore, the MGSA is tested on ten functions provided by CEC 2005 special session and compared with various modified Gravitational Search Algorithm, Particle Swarm Optimization, and Genetic Algorithm. The obtained results confirm the high performance of the MGSA in solving various problems in optimization.