Improving defect prediction via JIT and enhanced effort aware metrics

Context: Software testing is critical in the software development life cycle, ensuring the quality and reliability of products by detecting and rectifying defects before deployment. As software complexity increases, predictive models using historical metrics and advanced machine learning techniques play a pivotal role in identifying likely defective modules, optimizing resource allocation, and reducing costs. Aim: This Ph.D. thesis aims to investigate the prediction of software defects by exploring the application and effectiveness of effort-aware metrics (EAMs) in the domain of software quality assurance. It intends to augment defect prediction accuracy using Just-In-Time (JIT) data for immediate corrective measures and to introduce and validate the novel Normalized PofB metric (NPofB), which recalibrates EAMs for more realistic and practical software quality assessments. Method: The research method includes a comprehensive analysis encompassing two distinct but complementary studies. The first delves into optimizing defect prediction using real-time JIT data, while the second investigates the application and limitations of EAMs. The methodological approach includes a thorough study of how EAM is used, the creation of the NPofB metric, and the creation of the ACUME tool to make EAM computations more automatic, aiming for reliable and repeatable results. Results: The results reveal that the integration of multiple EAMs, coupled with the proposed normalization method, significantly improves the understanding and effectiveness of defect classifiers. Findings indicate that prior studies may have underestimated the value of classifiers in defect ranking. The research also demonstrates the practical utility of the ACUME tool in fostering better research practices through reproducibility and accuracy. Conclusion: In conclusion, this thesis adds to the fields of software testing and defect prediction by introducing NPofBs and evaluating multiple EAMs.Thereby, the thesis not only enhances the precision of predictions but also advocates for a more efficient software development process. Ultimately, it lays the groundwork for future research to build upon, with the aim of delivering higher-quality software products in a cost-effective manner.