Approval Prediction for Software Enhancement Report using Deep Neural Network

Abstract

Software applications obtain enhancement requests on a large scale to fulfil user requirements through a bug tracking system, which is an important application for keeping record of the bugs in the software development process. Conventionally, software developers used to manually check these requests. However, manual inspection of these requests turns out to be a boring, hectic and time-consuming activity. Therefore, an automatic prediction system is required which can reject or approve an enhancement report without manual check. Existing approaches target this problem through machine learning techniques such as Multinomial Naive Bayes and Support Vector Machine. In this work we utilize a deep learning based approach for approval prediction of enhancement reports that include the following steps: First, we preprocess each enhancement report using natural language processing techniques. Then we perform emotion analysis of each preprocessed enhancement reports using Senti4SD to compute its sentiment. Next we combine the summary and description of enhancement report as a sequence of words to learn the features with some attention mechanism using deep recurrent neural network (RNN). Finally, on the basis of features selection model and sentiment we train deep neural network based classifier, the convolutional neural network (CNN) to predict the approval prediction of enhancement report. We evaluate the proposed approach on an open-source bug tracking system. Results of numerical evaluation suggest that the proposed approach significantly outperforms the existing approaches and achieve the accuracy of 82.15%, precision 90.56%, recall 80.10% and f-measure as 85.01%.