Improved prediction accuracy with reduced feature set using novel binary gravitational search optimization

Improvement of classifier prediction accuracy is a long run burning issue over the years in the field of data mining and machine learning application. Optimized feature set is the best strategy and feature selection is the only key to the optimization problem. Various heuristic search algorithms are proposed in the literature for the feature set selection task. In this context we have enlightened the feature set exploration capacity of gravitational search algorithm (GSA) which is based on the Newton’s law of motion principle and the interaction of masses. Binary version of GSA with one modification is used for our application here. It is found that binary gravitational search algorithm (BGSA) is useful for finding only the relevant features while improving classifier accuracy from that with all features. We test our approach on six benchmark datasets from UCI machine learning repository.