Machine learning in software defect prediction: A business-driven systematic mapping study

Context: Machine learning is a valuable tool in software engineering allowing fair defect prediction capabilities at a relatively small expense. However, although the practical usage of machine learning in defect prediction has been studied over many years, there is not sufficient systematic effort to analyse its potential for business application.Objective: The following systematic mapping study aims to analyse the current state-of-the-art in terms of machine learning software defect prediction modelling and to identify and classify the emerging new trends. Notably, the analysis is done from a business perspective, evaluating the opportunities to adopt the latest techniques and methods in commercial settings to improve software quality and lower the cost of development life cycle.Method: We created a broad search universe to answer our research questions, performing an automated query through the Scopus database to identify relevant primary studies. Next, we evaluated all found studies using a classification scheme to map the extent of business adoption of machine learning software defect prediction based on the keywords used in the publications. Additionally, we use PRISMA 2020 guideline to validate reporting.Results: After the application of the selection criteria, the remaining 742 primary studies included in Scopus until February 23, 2022 were mapped to classify and structure the research area. The results confirm that the usage of commercial datasets is significantly smaller than the established datasets from NASA and open-source projects. However, we have also found meaningful emerging trends considering business needs in analysed studies.Conclusions: There is still a considerable amount of work to fully internalise business applicability in the field. Performed analysis has shown that purely academic considerations dominate in published research; however, there are also traces of in vivo results becoming more available. Notably, the created maps offer insight into future machine learning software defect prediction research opportunities.