Feature Selection Based on Genetic Algorithm With Stochastic Disturbance Local Optimization

Abstract

The paper proposes a feature selection method based on genetic algorithm with stochastic disturbance local optimization (GASD) for data dimension reduction problem. In this algorithm, a local search module is introduced into every search iteration under the global search framework of genetic algorithm. In the local search, a stochastic disturbance mechanism is utilized to update the current optimal feature subset. The optimal feature subset is obtained by using global search and optimized local search. Experimental results show that GASD can effectively delete redundant features, reduce data dimensions, and improve the generalization ability of classification model, especially in high-dimensional data.