Enhancing Software Defect Prediction: Exploring the Predictive Power of Two Data Flow Metrics

Data flow coverage criteria find extensive application in software testing, yet scant research exists regarding low-level data flow metrics as predictors of software defects. Objectives: Within this context, we explore two such metrics - dep-degree (DD), as proposed by Beyer and Fararooy, and a novel data flow metric named dep-degree density (DDD). Approach: We investigate the significance of DD and DDD within Software Defect Prediction (SDP) models. Through correlation analysis, we assess whether DD and DDD capture distinct code characteristics in comparison to established source code metrics related to size, complexity, and documentation. Lastly, we conduct experiments utilizing five different classifiers on nine projects from the Unified Bug Dataset, aiming to compare the performance of SDP models trained with and without data flow metrics. Results: 1) DD displays notable correlations with several other code metrics, while DDD either exhibits weak correlations or lacks correlation with other metrics under investigation; 2) both DD and DDD emerge with high rankings in the feature importance analysis; 3) SDP models incorporating DD and DDD outperform models excluding data flow metrics. Conclusions: Data-flow metrics, DD and DDD, hold the potential to serve as valuable predictors in SDP models.