A weighted bacterial colony optimization for feature selection

Abstract

Feature selection is essentially important for high dimensional feature characterization problems. In this paper, we propose a weighted feature selection algorithm based on bacterial colony optimization (BCO) for dimensionality reduction. The weighted strategy is used for reducing the redundant features as well as increasing the classification performance, which considers the frequency of features being selected by bacterial colony optimization(BCO) as well as the repeated appearance in the same individual. The contributions of features in classification will be evaluated and kept in 'Achieve'. The learning mechanism used in BCO considers the randomness which avoids the ignorance of unseen features as well as disengages from the local optimal error. Benchmark datasets with varying dimensionality are selected to test the effectiveness of the proposed feature selection method. The significance of the proposed weight feature selection algorithm is verified by comparing with three recently proposed population based feature selection algorithms. © 2014 Springer International Publishing Switzerland.