On the Costs and Profit of Software Defect Prediction

Defect prediction can be a powerful tool to guide the use of quality assurance resources. However, while lots research covered methods for defect as well methodological aspects research, actual cost saving potential is still unclear. Within this article, we close gap and formulate model software prediction. We derive mathematically provable boundary conditions that must fulfilled by models such there positive profit when used. Our includes like costs assurance, post-release defects, possibility fails reveal predicted relationship between artifacts defects. initialize using different assumptions, perform experiments show trends behavior on real projects. results unrealistic assumption defects only affect single artifact, which standard practice in literature, leads inaccurate estimations. Moreover, indicate thresholds machine learning metrics are also not suited define success criteria