Software Bug Prediction Model Based on Mathematical Graph Features Metrics

Quality assurance is one of the most important activities in software development and maintenance. Software source codes are modified via change requests, functional improvement, and refactoring. When software changes, it is difficult to define the scope of test cases, and software testing costs tend to increase to maintain software quality. Therefore, change analysis is a challenge, and static testing is a key solution to this challenge. In this study, we propose new static testing metrics using mathematical graph analysis techniques for the control flow graph generated from the three-address code of the implementation codes based on our hypothesis of the existing correlation between the graph features and any software bugs. Five graph features are strongly correlated with the software bugs. Hence, our bug prediction model exhibits a better performance of 0.25 FN, 0.04 TN ratio, and 0.08 precision than a model based on the traditional bug prediction metrics, which are complexity, line of code (steps), and CRUD.