Performance and cost-effectiveness of change burst metrics in predicting software faults

Analysis of Developers’ Network and Change Burst Metrics as Predictors of Software Faults

“…This research is an extension of a study carried out by Ndenga et al (2019). It extended this study by comparing performance of developers' network metrics to change burst metrics as predictors of software faults.…”

“…Similarly, Ndenga et al (2019) found out that change burst metrics models showed the highest fault detection probabilities ranging between 50% and 68% as compared to change, code churn, organizational and source code metrics which exhibited lower probabilities when 20% of code files were examined. Finally, they found out that change burst metrics models had the least cost of misclassification of components in comparison to change, code churn, organizational, and source code metrics for three out of four projects (Ndenga et al, 2019). Studies done by Ndenga et al (2019) and Nagappan et al (2010) are all concluding that change burst metrics are superior software fault predictors.…”

“…Finally, they found out that change burst metrics models had the least cost of misclassification of components in comparison to change, code churn, organizational, and source code metrics for three out of four projects (Ndenga et al, 2019). Studies done by Ndenga et al (2019) and Nagappan et al (2010) are all concluding that change burst metrics are superior software fault predictors. Unfortunately, none of the studies has analyzed the performance of change burst metrics to that of developers' network metrics on fault detection probability and the cost of misclassification of components within a common experimental context.…”

“…However according to Nguyen et al (2010), recall values registered by SNA measures are similar or worse than the recall values registered by complexity metrics when these measures are used as software fault predictors. Ndenga et al (2019) carried out experiments on Jmeter, gedit, POI and GIMP projects and found out that change burst metrics models registered superior performances for almost all numerical performance measures as compared to change, code churn, organizational and source code metrics. Similarly, Ndenga et al (2019) found out that change burst metrics models showed the highest fault detection probabilities ranging between 50% and 68% as compared to change, code churn, organizational and source code metrics which exhibited lower probabilities when 20% of code files were examined.…”

“…Ndenga et al (2019) carried out experiments on Jmeter, gedit, POI and GIMP projects and found out that change burst metrics models registered superior performances for almost all numerical performance measures as compared to change, code churn, organizational and source code metrics. Similarly, Ndenga et al (2019) found out that change burst metrics models showed the highest fault detection probabilities ranging between 50% and 68% as compared to change, code churn, organizational and source code metrics which exhibited lower probabilities when 20% of code files were examined. Finally, they found out that change burst metrics models had the least cost of misclassification of components in comparison to change, code churn, organizational, and source code metrics for three out of four projects (Ndenga et al, 2019).…”