Enhanced Crow Search Algorithm for Feature Selection

Abstract The crow search algorithm (CSA) is a recent metaheuristic inspired by the intelligent group behavior of crows. It has attracted the attention of many researchers because of its simplicity and easy implementation. However, it suffers from premature convergence because of its ability to balance between exploration and exploitation is weak. Therefore, we investigate in this paper, an enhanced version of CSA called by us ECSA as a wrapper feature selection method to extract the best feature subsets. This enhancement achieved by introducing three modifications to the original CSA to improve its performance. Firstly, we propose an adaptive awareness probability to enhance the balance between exploration and exploitation. Secondly, we replace the random choice of the crow to follow by the dynamic local neighborhood to guide the local search. Thirdly, we introduce a novel global search strategy to increase the global exploration capability of the crow. The performance of ECSA is measured using three performance metrics and statistical significance over 16 datasets from the UCI repository. The obtained results are compared with those of the original CSA and some state-of-the-art techniques in the literature. Experimental results showed that ECSA presents a better convergence speed and a better-quality solution.