Severity Prediction for Bug Reports using Tree-based Ensemble Models: A Comparative Study

Predicting the severity of software bug reports is an important research issue. In recent studies, ensemble approaches have been employed in various prediction tasks for bug reports to obtain performance improvements. In this study, we investigate the prediction performance of four tree-based ensemble models on the severity prediction task. The studied ensemble models include Random Forest, Extra Trees, XGBoost, and LightGBM. In addition, particle swarm optimization (PSO) is used to find the suitable weights of different features for severity prediction. The prediction performance is evaluated with datasets of three open- source projects, Eclipse, Mozilla, and Gentoo Linux. The experimental results show that LightGBM with PSO-optimized feature weights can achieve the best prediction performance in many performance metrics.