Abstract
Software defect prediction (SDP) has been a long-standing field of research in software engineering. It is both essential and challenging in software development projects. Moreover, it plays a critical role in improving software quality. SDP methods are primarily based on software code and software code metrics, and generally employ machine learning techniques. Therefore, these methods are typically applied during or after the development phase. In this study, we propose an exploratory, correlation-based novel method for predicting code defect-proneness at the requirements analysis phase by considering COSMIC Function Point (CFP) measurements for the first time. An industrial case study was performed in the context of a FinTech company. We measured CFP during the requirements analysis phase of the corresponding project for all requirements over a 12-month period. Additionally, we collected the code defects identified during the testing phase and also the seniority levels of developers. Our results show that the number of code defects in software varies depending on factors such as the experience of developers, the platform on which the development is carried out, and the CFP data movements. Moreover, there is a correlation between the identified factors and the number of code defects. By using these results, it is possible to perform exploratory prediction of code defect-proneness from the requirements before development even begins. Furthermore, these outcomes can be utilized to guide development and test engineers, thereby enhancing their awareness and focus during the development and testing processes.
Available at https://doi.org/10.1007/s11219-025-09736-1
