摘要：The paper presents an analysis of 83 versions of industrial, open-source and academic projects. Wehave empirically evaluated whether those project types constitute separate classes of projects withregard to defect prediction. Statistical tests proved that there exist significant differences betweenthe models trained on the aforementioned project classes. This work makes the next step towardscross-project reusability of defect prediction models and facilitates their adoption, which has beenvery limited so far.
