An efficient feature subset selection approach for machine learning

Abstract

This literature introduces CAPPER, an efficient hybrid feature subset selection approach which is the combination of feature subsets from the Correlation-based feature selection (CFS) and Wrapper approaches. CFS is basically a filter approach that appraises the merits of subset attributes by classifying the feature ability according to the amount of redundancy between them and the feature subset selection for Wrappers that examines the attributes by applying the induction of various machine learning algorithms. For the evaluation of metrics, the CAPPER approach is tested on the different domains of datasets. The reduced, highly merit and accurate feature subsets obtained from CAPPER approach are then trained with the machine learning algorithm and evaluated by cross-validation for the set of attributes. Moreover, a statistical approach is applied for the significance of the result. It was observed that the CAPPER approach surpasses the CFS and Wrapper approaches on different domains of datasets.