An Ensemble Of Feature Selection Algorithms Using OWA Operator

Abstract

This article tries to deal with the problem of ensemble feature selection as a rank aggregation procedure. As our proposed algorithm, EFS-OWA first constructs a decision matrix based on the rank of features obtained by different feature selection algorithms. To aggregate the results of different feature selection methods, we used Ordered Weighted Averaging (OWA) aggregation operator. This will allow the features that have the most satisfactory by feature selection algorithms assigned highest ranks. Some ensemble feature selection algorithms that use rank aggregation procedures are compared with EFS-OWA to prove the performance of the proposed method. Also, another comparison is conducted between EFS-OWA and basic feature selection algorithms. All these comparisons are made based on classification accuracy and the runtime of algorithms. Based on the experiment results, we can see that EFS-OWA outperforms competitive methods.