Fuzzy set theoretic measures for automatic feature evaluation: II

Abstract

The present paper is a continuation of our previous work [1] in which we developed an algorithm for automatic ranking of the individual feature importance for pattern recognition problems. Here we have extended the work by 1. (i) evaluating the importance of any subset of features collectively 2. (ii) providing an average feature evaluation index considering all the classes 3. (iii) comparing the results with those of statistical measures considering their variation with interest distance. Effectiveness of the algorithm is demonstrated on six-class, three-feature vowel data; four-class, five-feature consonant data; and three-class fifteen-feature mango leaf data.