Filter � GA Based Approach to Feature Selection for Classification

Abstract

This paper presents a new approach to select reduced number of features in databases. Every database has a given number of features but it is observed that some of these features can be redundant and can be harmful as well as and can confuse the process of classification. The proposed method applies filter attribute measure and binary coded Genetic Algorithm to select a small subset of features. The importance of these features is judged by applying K-nearest neighbor (KNN) method of classification. The best reduced subset of features which has high classification accuracy on given databases is adopted. The classification accuracy obtained by proposed method is compared with that reported recently in publications on twenty eight databases. It is noted that proposed method performs satisfactory on these databases and achieves higher classification accuracy but with smaller number of features.