Bug Severity Prediction Algorithm Using Topic-Based Feature Selection and CNN-LSTM Algorithm

Increasing software usage has gradually increased the occurrence of bugs. When writing a bug report, severity can be freely selected, so subjective judgment author is involved. In judgment, error may occur depending on background knowledge between user and developer. To resolve this problem, in paper, was predicted using feature selection algorithm each topic. We utilize dataset Eclipse Mozilla open source projects. First, we classify reports by topic-based severity, extract features from The learning characteristics CNN-LSTM algorithm, F-measure 90.62% 93.22% Mozilla. evaluate effectiveness proposed model, compared baselines including DeepSeverity EWD-Multinomial studies with projects showed that model more efficient.