Improving Evolutionary Algorithm Performance for Feature Selection in High-Dimensional Data

In classification and clustering problems, selecting a subset of discriminative features is a challenging problem, especially when hundreds or thousands of features are involved. In this framework, Evolutionary Computation (EC) techniques have received a growing scientific interest in the last years, because they are able to explore large search spaces without requiring any a priori knowledge or assumption on the considered domain. Following this line of thought, we developed a novel strategy to improve the performance of EC-based algorithms for feature selection. The proposed strategy requires to rank the whole set of available features according to a univariate evaluation function; then the search space represented by the first M ranked features is searched using an evolutionary algorithm for finding feature subsets with high discriminative power. Results of comparisons demonstrated the effectiveness of the proposed approach in improving the performance obtainable with three effective and widely used EC-based algorithm for feature selection in high dimensional data problems, namely Ant Colony Optimization (ACO), Particle Swarm Optimization (PSO) and Artificial Bees Colony (ABC).