APPLICATION OF XGBOOST ALGORITHM AND DEEP LEARNING TECHNIQUES FOR SEVERITY ASSESSMENT OF SOFTWARE DEFECT REPORTS

Abstract

Software is present in every aspect of our everyday life, and defects are bound to be found during the testing of the software, no matter how small. It is therefore imperative for software testing engineers to assess the severity of software defects to allocate proper resources for the correction of the defects and prevent software crashes. In this study, we have proposed the use of the Extreme Gradient Boosting Technique and deep learning techniques: Convolutional Neural Network and Recurrent Neural Network to predict the severity of the defects occurring in the software. AUC and sensitivity are the metrics used to evaluate the results. All three techniques: XGBoost algorithm, CNN and RNN have performed really well in predicting the severities for all the defects. It has also been noted that XGBoost algorithm is the most efficient in predicting high severity defects, while the performance of deep learning techniques is excellent for the highest as well as the lowest severity defects. For the rest of the severity values, the performance of all the three classifiers is fairly consistent.