A comparison between the Pittsburgh and Michigan approaches for the binary PSO algorithm

Abstract

This paper shows the performance of the binary PSO algorithm as a classification system. These systems are classified in two different perspectives: the Pittsburgh and the Michigan approaches. In order to implement the Michigan approach binary PSO algorithm, the standard PSO dynamic equations are modified, introducing a repulsive force to favor particle competition. A dynamic neighborhood, adapted to classification problems, is also defined. Both classifiers are tested using a reference set of problems, where both classifiers achieve better performance than many classification techniques. The Michigan PSO classifier shows clear advantages over the Pittsburgh one both in terms of success rate and speed. The Michigan PSO can also be generalized to the continuous version of the PSO.