Correlation Feature Selection Weighting Algorithms for Better Support Vector Classification: An Empirical Study

Abstract

Characteristics of Support Vector Machine (SVM) and its classifications are elaborated to show why incorporation of newly proposed and formulated regularization on feature selections based on correlation studies are necessary to achieve a better prediction or classification. Feature selections based  on correlation studies are incorporated into the proposed formulations for the weighting portions of the objective functions for SVM. Proposed cfsw-SVM algorithms arethen developed. Proposed formulations on SVM regularization parameter provides synergistic adjustments between prediction or classification accuracy and the level of correlations among features in the SVM implemented. Prediction and/or classification accuracies of cfsw-SVM algorithms are significantly improved.