DRS: A Developer Risk Metric for Better Predicting Software Fault-Proneness

Previous studies have reported that the performance of a developer can greatly impact the quality of the software he/she has worked on. Such performance can be measured using two developer risk metrics during a particular development period. One is the ratio of the number of bug-introduce commits to the total number of commits made by a developer (i.e., the DQ metric). The other is the proportion of faulty software modules out of all modules modified by the developer (i.e., the BR metric). However, all bug-introduce commits, no matter its severity, are treated equally by both DQ and BR metrics. Moreover, the complexity of a software module that a developer is working on may also have a potential impact on his/her performance but is not considered by either DQ or BR. To resolve these two problems, we propose Developer Risk Score (DRS), which takes both program complexity and the severity of bug-introduce commits into account, to evaluate the performance of a developer. Nine software risk metrics based on DRS are further derived to predict the fault proneness of a given software module. Results from our case studies show that (1) DRS-based software risk metrics are generally more correlated with the number of bugs in a software module and the cumulative severity score of bug-introduce commits for a module than DQ-and BR-based metrics, and (2) models using DRS-based metrics are generally more effective in predicting software fault-proneness than those using DQ-and BR-based metrics.