Issues with SZZ: An empirical assessment of the state of practice of defect prediction data collection
share
Defect prediction research has a strong reliance on published data sets that are shared between researchers. The SZZ algorithm is the de facto standard for collecting defect labels for this kind of data and is used by most public data sets. Thus, problems with the SZZ algorithm may have a strong indirect impact on almost the complete state of the art of defect prediction. Recent research uncovered potential problems in different parts of the SZZ algorithm. Within this article, we provide an extensive empirical analysis of the defect labels created with the SZZ algorithm. We used a combination of manual validation and adopted or improved heuristics for the collection of defect data to establish ground truth data for bug fixing commits, improved the heuristic for the identification of inducing changes for defects, as well as the assignment of bugs to releases. We conducted an empirical study on 398 releases of 38 Apache projects and found that only half of the bug fixing commits determined by SZZ are actually bug fixing. Moreover, if a six month time frame is used in combination with SZZ to determine which bugs affect a release, one file is incorrectly labeled as defective for every file that is correctly labeled as defective. In addition, two defective files are missed. We also explored the impact of the relatively small set of features that are available in most defect prediction data sets, as there are multiple publications that indicate that, e.g., churn related features are important for defect prediction. We found that the difference of using more features is negligible.
READ FULL TEXT