Ant colony optimization with null heuristic factor for feature selection

Recently, the ant colony optimization (ACO) meta-heuristic has received more attention as an efficient searching method for feature selection. This paper addresses various solution representation schemes of ACO and their effectiveness with respect to whether they consider correlations between features. A generic code of ACO using on-edge representation is presented. The paper formulates the ¿-component by concentrating on the types of objects that participate in calculating the ¿ value. Four schemes based on the formulation are compared in terms of the timing efficiency and accuracy. The experimental results showed that the null-¿ scheme is comparable to other schemes. We discuss the explanation of these conclusions.