Ant Colony Optimization for Rule Induction with Simulated Annealing for Terms Selection

This paper proposes a sequential covering based algorithm that uses an ant colony optimization algorithm to directly extract classification rules from the data set. The proposed algorithm uses a Simulated Annealing algorithm to optimize terms selection, while growing a rule. The proposed algorithm minimizes the problem of a low quality discovered rule by an ant in a colony, where the rule discovered by an ant is not the best quality rule, by optimizing the terms selection in rule construction. Seventeen data sets which consist of discrete and continuous data from a UCI repository are used to evaluate the performance of the proposed algorithm. Promising results are obtained when compared to the Ant-Miner algorithm and PART algorithm in terms of average predictive accuracy of the discovered classification rules.