Entropy Based Machine Learning Models for Software Bug Severity Assessment in Cross Project Context

There can be noise and uncertainty in the bug reports data as the bugs are reported by a heterogeneous group of users working across different countries. Bug description is an essential attribute that helps to predict other bug attributes, such as severity, priority, and time fixes. We need to consider the noise and confusion present in the text of the bug report, as it can impact the output of different machine learning techniques. Shannon entropy has been used in this paper to calculate summary uncertainty about the bug. Bug severity attribute tells about the type of impact the bug has on the functionality of the software. Correct bug severity estimation allows scheduling and repair bugs and hence help in resource and effort utilization. To predict the severity of the bug we need software project historical data to train the classifier. These training data are not always available in particular for new software projects. The solution which is called cross project prediction is to use the training data from other projects. Using bug priority, summary weight and summary entropy, we have proposed cross project bug severity assessment models. Results for proposed summary entropy based approach for bug severity prediction in cross project context show improved performance of the Accuracy and F-measure up to 70.23% and 93.72% respectively across all the machine learning techniques over existing work.