Hybrid Multiple Deep Learning Models to Boost Blocking Bug Prediction

Hybrid Multiple Deep Learning Models to Boost Blocking Bug Prediction

A blocking bug (BB) is a severe bug that could prevent other bugs from being fixed in time and cost more effort to repair itself in software maintenance. Hence, early detection of BBs is essential to save time and labor costs. However, BBs only occupy a small proportion of all bugs during software life cycle, making it difficult for developers to identify these blocking relationships. This study proposes a novel blocking bug prediction approach based on the hybrid deep learning model, a combination of Bi-directional Long Short-Term Memory (Bi-LSTM) and Convolutional Neural Network (CNN). Our approach first extracts summaries and descriptions from bug reports to construct two classifiers, respectively. Second, our approach combines two classifiers into a hybrid model to predict the blocking relationship of each blocking bug pair. Finally, our approach produces a report of identified blocking bugs for developers. To investigate the effectiveness of proposed approach, we conducted an empirical study on bug reports of seven large-scale projects. The final experimental results illustrate that our approach can perform better than the recent state-of-the-art baselines. Precisely, the hybrid model can predict BB better with an average accuracy of 57.20%, and an improvement of 73.53% in terms of the F1-measure when compared to ELBlocker. Moreover, according to the bug report’s description, BB can be predicted well with an average accuracy of 49.16%.