Bee Colony Optimization for Multi-Label Feature Selection

Abstract

In this paper, we consider the problem of feature selection for multi-label data. Multi-label feature selection is a process of finding the appropriate subset of features that allows multi-label classifiers to find better solutions in a shorter amount of time. For this purpose, we developed the Bee Colony Optimization algorithm based on mutual information and compared it with other metaheuristics from literature, i.e. Ant Colony Optimization and Memetic Algorithm. After testing it on several benchmark instances, we concluded that our approach outperforms the other two methods.