Early Identification of Invalid Bug Reports in Industrial Settings - A Case Study

Software development companies spend considerable time resolving bug reports. However, bug reports might be invalid, i.e., not point to a valid flaw. Expensive resources and time might be expended on invalid bug reports before discovering that they are invalid. In this case study, we explore the impact of invalid bug reports and develop and assess the use of machine learning (ML) to indicate whether a bug report is likely invalid. We found that about 15% of bug reports at the case company are invalid, and that their resolution time is similar to valid bug reports. Among the ML-based techniques we used, logistic regression and SVM show promising results. In the feedback, practitioners indicated an interest in using the tool to identify invalid bug reports at early stages. However, they emphasized the need to improve the explainability of ML-based recommendations and to reduce the maintenance cost of the tool.

Automated summary

This study investigates the impact of invalid bug reports and explores the use of machine learning to identify them. It is found that 15% of bug reports are invalid, and logistic regression and SVM show promising results in identifying them.