Abductive Network Ensembles for Improved Prediction of Future Change-Prone Classes in Object-Oriented Software

Software systems are subject to a series of changes due to a variety of maintenance goals. Some parts of the software system are more prone to changes than others. These change-prone parts need to be identified so that maintenance resources can be allocated effectively. This paper proposes the use of Group Method of Data Handling (GMDH)-based abductive networks for modeling and predicting change proneness of classes in object-oriented software using both software structural properties (quantified by the C&K metrics) and software change history (quantified by a set of evolution-based metrics) as predictors. The empirical results derived from an experiment conducted on a case study of an open-source system show that the proposed approach improves the prediction accuracy as compared to statistical-based prediction models.