Novel Mutation Operators of a Variable-Length Representation for EC-Based Feature Selection in High-Dimensional Data

Abstract

Because of their ability to explore large and complex search spaces, evolutionary algorithms have been widely used to solve feature selection problems. Many of them, e.g. genetic algorithms, provide a natural way to represent feature subsets, however, when the number of features is high, these representations may make the evolutionary process inefficient. In a previous paper, we presented a novel variable-length representation scheme for encoding subsets of features efficiently as well as a crossover operator to cope with the variable length representation. In this paper, we present two novel mutation operators that can generate offspring that may have a length different from that of the parent. We aim to simplify the approach presented previously, by using a single operator. We have tested the proposed approach on six datasets and the results compared with those achieved by a standard GA and two state-of-art algorithms. The results of comparisons demonstrated the effectiveness of the proposed approach when thousand of features are involved.