A comparative analysis of evolutionary algorithms for the prediction of software change

Change-proneness prediction of software components has become a significant research area wherein the quest for the best classifier still persists. Although numerous statistical and Machine Learning (ML) techniques have been presented and employed in the past literature for an efficient generation of change-proneness prediction models, evolutionary algorithms, on the other hand, remain vastly unexamined and unaddressed for this purpose. Bearing this in mind, this research work targets to probe the potency of six evolutionary algorithms for developing such change prediction models, specifically for source code files. We employ apposite object oriented metrics to construct four software datasets from four consecutive releases of a software project. Furthermore, the prediction capability of the selected evolutionary algorithms is evaluated, ranked and compared against two statistical classifiers using the Wilcoxon signed rank test and Friedman statistical test. On the basis of the results obtained from the experiments conducted in this article, it can be ascertained that the evolutionary algorithms possess a capability for predicting change-prone files with high accuracies, sometimes even higher than the selected statistical classifiers.