Enhancing Defect Prediction with Static Defect Analysis

Abstract

In the software development process, how to develop better software at lower cost has been a major issue of concern. One way that helps is to find more defects as early as possible, on which defect prediction can provide effective guidance. The most popular defect prediction technique is to build defect prediction models based on machine learning. To improve the performance of defect prediction model, selecting appropriate features is critical. On the other hand, static analysis is usually used in defect detection. As static defect analyzers detects defects by matching some well-defined "defect patterns", its result is useful for locating defects. However, defect prediction and static defect analysis are supposed to be two parallel areas due to the differences in research motivation, solution and granularity. In this paper, we present a possible approach to improve the performance of defect prediction with the help of static analysis techniques. Specifically, we present to extract features based on defect patterns from static defect analyzers to improve the performance of defect prediction models. Based on this approach, we implemented a defect prediction tool and set up experiments to measure the effect of the features.