A NEW NON-PARAMETRIC FEATURE SELECTION CRITERION-EFFECT

Abstract

The problem of feature selection in the field of pattern classification of multidimensional feature vectors is essentially one of dimensionality reduction. A new non-parametric technique for choosing an optimum subset of features from the given feature set is proposed in this study. This technique is based on the concept of inter class and intra-class distances and tests conducted reveal tile efficacy of this new Effective Figure of Merit Criterion 'EFFECT'.