A MODIFIED T-SCORE FOR FEATURE SELECTION

Abstract

In this study, an alternative approach to t-score method, one of the feature selection methods, has been suggested and some analyses have been executed in order to compare t-score method and our approach. When comparing them, commonly used data sets in data mining studies, Arcene, Gisette and Madelon have been used. In line with the purpose of this study, the first 50, 100, 150 and 200 features for each data set has been selected, in consequence, 24 data subsets have been created. The classification accuracies of t-score and suggested method has been compared by using these data subsets. When calculating the classification accuracies, two commonly used methods in literature, Artificial Neural Networks and Support Vector Machines have been used. According to this study, the result of the suggested feature selection method is statistically more successful than t-score.