Automatic prediction of the severity of bugs using stack traces
Korosh Koochekian Sabor, Mohammad Hamdaqa, Abdelwahab Hamou‐Lhadj · PolyPublie (École Polytechnique de Montréal) · 2016
The severity of a bug is a measure of how a defect affects the functionality of a system. Developers refer to the severity of the reported bugs to prioritize the handling of bug reports. The process of assigning a severity level to a bug is performed manually, often by inexperienced users, making it time consuming and error prone. Existing techniques for automatically predicting the severity of bugs rely on text mining and information retrieval algorithms applied to the description of the bugs. The problem is that the description tends to be too informal and not quite reliable. In this paper, we show how information found in stack traces (a more formal source of data containing the history of function calls to the function in which the crash happened) can be used to automatically predict the severity of bugs. Our experiments with Eclipse bug reports submitted between 2001 to 2015 show that stack traces are a better feature for predicting the severity of bugs than the bug description.