Predicting software maintainability using ensemble techniques and stacked generalization

The prediction of software maintainability has emerged as an important research topic to address industry expectations for reducing costs, in particular maintenance costs. In the last decades, many studies have used single techniques to predict software maintainability but there is no agreement as to which technique can achieve the best prediction. Ensemble techniques, which combine two or more techniques, have been investigated in recent years. This study investigates ensemble techniques (homogeneous as well as heterogeneous) for predicting maintainability in terms of line code changes. To this end, well-known homogeneous ensembles such as Bagging, Boosting, Extra Trees, Gradient Boosting, and Random Forest are investigated first. Then the stacked generalization method is used to construct heterogeneous ensembles by combining the most accurate ones per dataset. The empirical results suggest that Gradient Boosting and Extra Trees are the best ensembles for all datasets, since they ranked first and second, respectively. Moreover, the findings of the evaluation of heterogeneous ensembles constructed using stacked generalization showed that they gave better prediction accuracy compared to all homogeneous ensembles. Copyright © 2020 for this paper by its authors.