Focusing on rule quality and pheromone evaporation to improve ACO rule mining

Abstract

In this paper an improved version of Ant-Miner algorithm is introduced and compared to the previously proposed ant-based rule mining algorithms. Our algorithm modifies the rule pruning process and introduces a dynamic pheromone evaporation strategy. The algorithm was run on five standard datasets and the average accuracy rate and numbers of discovered rules were analyzed as two important performance metrics of rule mining. As simulation results show, not only the accuracy rate and rule comprehensiveness is improved by our algorithm, the algorithm runtime is also reduced.