An improved Michigan particle swarm optimization for classification

Classification is one of the most frequently occurring tasks of human decision making. In this paper, two improved versions of Michigan particle swarm optimization (MPSO), Improved MPSO1 (IMPSO1) and Improved MPSO2 (IMPSO2), are proposed. IMPSO1 adopts a adaptive inertia factor so as to flexibly control the search path, moreover, both nearest neighbor (NN) and 5-NN classification are used so as to take more local information into account and improve the diversity of population. In IMPSO2, a new selection operator is introduced into MPSO to obtain a competitive classification success rate as well as a lower computation cost. The proposed algorithm has been extensively compared with PSO, MPSO, C4.5, 1-NN and 3-NN over eight UCI data sets. The result of experiment indicates the superiority of the algorithm over other four algorithms on classification success rate.