Software Change Prediction: A Systematic Review and Future Guidelines

Abstract

Background: The importance of Software Change Prediction (SCP) has been emphasized by several studies. Numerous prediction models in literature claim to effectively predict change-prone classes in software products. These models help software managers in optimizing resource usage and in developing good quality, easily maintainable products. Aim: There is an urgent need to compare and assess these numerous SCP models in order to evaluate their effectiveness. Moreover, one also needs to assess the advancements and pitfalls in the domain of SCP to guide researchers and practitioners. Method: In order to fulfill the above stated aims, we conduct an extensive literature review of 38 primary SCP studies from January 2000 to June 2019. Results: The review analyzes the different set of predictors, experimental settings, data analysis techniques, statistical tests and the threats involved in the studies, which develop SCP models. Conclusion: Besides, the review also provides future guidelines to researchers in the SCP domain, some of which include exploring methods for dealing with imbalanced training data, evaluation of search-based algorithms and ensemble of algorithms for SCP amongst others.

Keywords

• change-proneness
• machine learning
• software quality
• systematic review