Classification rule discovery with ant colony optimization algorithm

Abstract

Ant Colony Algorithms has been successfully applied to solve combinatorial optimization problems. Subsequently applications on Data Mining (DM) appeared, more specifically aiming to solve classification problems. The Ant-Miner [3] algorithm is a good example of a solution to this problem. This algorithm is better than C4.5 [7] and CN2 [8]. This paper presents a new algorithm which applies an innovative modeling of the foraging behavior of ants [4] to the Ant-Miner. As a result of this adaptation, four different versions of the Ant-Miner algorithm were generated, tested and compared to the original version using seven public domain data sets. One of the versions produced comparatively superior results in terms of predictive accuracy in different system configurations. © 2012 Springer-Verlag.