A high-dimensional feature selection method based on modified Gray Wolf Optimization

For data mining tasks on high-dimensional data, feature selection is a necessary pre-processing stage that plays an important role in removing redundant or irrelevant features and improving classifier performance. The Gray Wolf optimization algorithm is a global search mechanism with promising applications in feature selection, but tends to stagnate in high-dimensional problems with locally optimal solutions. In this paper, a modified gray wolf optimization algorithm is proposed for feature selection of high-dimensional data. The algorithm introduces ReliefF algorithm and Coupla entropy in the initialization process, which effectively improves the quality of the initial population. In addition, modified gray wolf optimization includes two new search strategies: first, a competitive guidance strategy is proposed to update individual positions, which make the algorithm’s search more flexible; second, a differential evolution-based leader wolf enhancement strategy is proposed to find a better position where the leader wolf may exist and replace it, which can prevent the algorithm from falling into local optimum. The results on 10 high-dimensional small-sample gene expression datasets demonstrate that the proposed algorithm selects less than 0.67% of the features, improves the classification accuracy while further reducing the number of features, and obtains very competitive results compared with some advanced feature selection methods. The comprehensive study analysis shows that proposed algorithm better balances the exploration and exploration balance, and the two search strategies are conducive to the improvement of gray wolf optimization search capability.