Software change prediction using voting particle swarm optimization based ensemble classifier

Prediction of change prone classes of a software has become an important area of research where the search for the best classifier still continues. While searching for an effective classifier, it needs to be ascertained whether an ensemble of classifier is better than its corresponding constituent classifiers. In this work, we propose four voting ensemble of classifiers, where a group of classifiers learn together and are used to create a single prediction model. The set of Particle Swarm Optimization (PSO) based classifiers are created based on five different fitness functions. Then the voting method is used to combine the predictions of these multiple classifiers so that the resultant model has improved accuracy. This proposal is based on the premises that while using a search-based algorithm for classification tasks, it is crucial to combine various classifiers based on different fitness function. The results of the study are statistically assessed on five popular Android application packages and advocate the use of a weighted voting ensemble of classifiers for developing change prediction models.¹