Bug summary entropy based training candidates identification in cross project severity prediction

In a software, prediction of bug severity helps in allocation of resources and scheduling during bug fixing. Automated bug severity prediction is required as assigning severity to a bug manually may take more time and sometimes severe bugs have to wait for a longer duration for fixing. A bug has different types of impact on the functionality and hence, resulted in different severity levels. To assess the severity level of bugs, different machine learning classifiers has been built on available project data. Bug severity automation process requires the available data of the project for training the classifier. For new and small software projects, the historical data may not be available in significant number for model training. In such situation, when data is not available for training the model, we can train the model with bug report data of similar project. About 63 training candidates have been designed by combining seven datasets of Eclipse projects to develop the severity prediction models. The information users submit regarding bugs during bug reporting may not be accurate because many users may not have the correct software knowledge. Also, the possibility of getting uncertainty during the process. The uncertainty in the data of bug report can be handled by using measures based on entropy. Handling of uncertainty during bug severity assessment has been carried out in the literature for the projects where sufficient number of data available for training the models. In this paper, an entropy based bug severity assessment model has been proposed with cross project prediction where models are trained and tested on different projects. We have developed 56 training candidates and each project has been tested against 35 training candidate(excluding the testing project). “Accuracy” and “F-measure” are the performance metrics that have been used to assess the effectiveness of the proposed approach. The proposed model has compared with the existing work available in the literature and significant improvement has been observed.