CNN based software bug severity prediction
Sweety Ahlawat, Dhiraj Khurana · 2025
In order to prioritize issues according to their relevance, teams that use software defect management are required to grade the severity of defects. This is because old manual classification procedures become more prone to subjectivity, inefficiency, and mistakes as software systems get more complex. This research provides an automated technique for forecasting the severity of defects that is based on convolutional neural networks (CNNs). The purpose of this paper is to solve the difficulties that have been raised. CNNs are able to discover complicated patterns and produce reliable severity estimates by evaluating textual data generated by bug reports. This is accomplished with little feature engineering by CNNs. As a result of using bug reports from the Mozilla and Eclipse projects, the suggested approach is assessed, and it achieves an accuracy of 90.6% and 91.3%, respectively. The accuracy, recall, and F-measure are three additional metrics that can be utilized to evaluate the efficacy of the model. According to the findings of the research, CNNs have the potential to greatly improve bug discovery.