Classification Rule Discovery with Ant Colony Optimization

In [9], we presented a modified version of Ant-Miner (i.e. Ant-Miner2), where the core computation heuristic value was based on a simple density estimation heuristic. In this paper, we present a further study and introduce another ant-based algorithm, which uses a different pheromone updating strategy and state transition rule. By comparison with the work of Parpinelli et al, our method can improve the accuracy of rule lists. Abstract—Ant-based algorithms or ant colony optimization (ACO) algorithms have been applied successfully to combinatorial optimization problems. More recently, Parpinelli and colleagues applied ACO to data mining classification problems, where they introduced a classification algorithm called Ant_Miner. In this paper, we present an improvement to Ant_Miner (we call it Ant_Miner3). The proposed version was tested on two standard problems and performed better than the original Ant_Miner algorithm. The remainder of the paper is organized as follow. In section 1, we present the basic idea of the ant colony systems. In section 2, the Ant_Miner algorithm (Rafael S.Parpinelli et al, 2000) is introduced. In section 3, the density based Ant_miner2 is explained. In section 4, our further improved method (i.e.Ant_Miner3) is shown. Then the computational results are reported in section 5. Finally, we conclude with general remarks on this work and further directions for future research.