Enhancing Bug-Fixing Time Prediction with LSTM-Based Approach

This work presents an approach based on Long short-term memory (LSTM) for estimating the bug-fixing time in the bug triage process. Existing bug-fixing time predictor approaches underutilize useful semantic information and long-term dependencies between activities in the bug-fixing sequence. Therefore, the proposed approach is a deep learning-based model that converts activities into vectors of real numbers based on their semantic meaning. It then uses LSTM to identify long-term dependencies between activities and classifies sequences as having either short fixing time or long fixing time. The evaluation on bug reports from the Eclipse project shows that this approach performs slightly better than the current best in the literature, boasting improved metrics such as accuracy, precision, f-score, and recall.