Prediction of software defect centralization using ai-based pathology

Abstract Software defect centralization is proposed as a metaphor to present the accumulation of various software degradation. It can describe the survival status of projects from the perspective of internal quality and help to fill the gap between quantitative fine-grained drawbacks and qualitative failure of projects. The current work on the detection and repair of software degradation cannot explain why some projects failed unexpectedly. Inspired by AI-based pathology, we define software defect centralization as a certain size of code that is growing and design a supervised prediction model. We build two datasets with 480 open-source Java projects to accomplish the model learning and evaluation. The experimental results show that the prediction model achieves an overall accuracy of 91.7% on the self-built test set. We also find that the degradation of bug-proneness and specific code smells are positively related to the presence of software defect centralization.