用机器学习自动预测软件缺陷严重程度，提升修复优先级效率。
Bug Severity Prediction in Software Projects Using Supervised Machine Learning Models
- 基于历史数据训练多种监督学习模型，融合文本特征与不平衡处理方法。
- 集成树模型和DistilBERT在整体准确率上表现最佳，线性模型更擅长发现关键缺陷。
- 为自动化缺陷分类提供算法选择依据，适合关注软件质量的开发团队。
缺陷严重程度预测在软件维护中至关重要，有助于开发团队优先处理对系统运行、稳定性和安全性有重大影响的缺陷。大型项目中缺陷库增长迅速，手动分类既费力又易受人为偏见影响。本文使用Eclipse Bugzilla的历史数据，对比了多种基于监督学习的分类器在缺陷严重程度预测上的表现，涵盖线性模型、梯度提升树、距离方法及基于Transformer的模型，以及从分词、TF-IDF、n-gram提取的文本特征与不平衡修正方法。评估指标包括准确率、精确率、召回率、F1值、AUC-ROC和混淆矩阵。结果表明，集成树方法和DistilBERT在整体准确率上领先，而线性模型在关键缺陷召回率上最优，反映出严重程度预测中的精度-召回权衡。研究为自动化缺陷分类提供了可操作的算法选型建议，有助于通过高效可扩展的优先级管理提升软件质量。
原文摘要 · Abstract (English)
Bug severity prediction is important in software maintenance, because it helps the development teams to prioritize bugs that have a significant impact on the operation, stability and security of the system. In large software projects bug repositories will grow at very rapid rate making classification of severity manual work labourious and unreliable and prone to human biasness. Many efforts have thus been dedicated on automated ways of severity prediction in the literature of software engineering research.This study compares different classifiers that are based on supervised machine learning algorithms for predicting bug severity levels using historical repository data from Eclipse Bugzilla. Evaluated methods range from linear classifiers, gradient boosting trees, distance method and transformer-based models, and text features, which are obtained from tokenization, TF-IDF, and n-grams and imbalance correction methods. Models were evaluated in terms of accuracy, precision, recall, F1 score, (AUC-ROC) and confusion matrix. Ensemble tree methods and DistilBERT achieved the top overall accuracy, while linear models performed best in recall of critical bugs which indicates some precision-recall tradeoff in imbalanced severity prediction. These findings provide useful actionable insight in choosing algorithms for automated bug triage, which can improve the quality of software through effective scalable prioritization. Keywords: Bug severity prediction, Supervised machine learning, Classification models, Software quality assurance, Historical bug data, Predictive analytics.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。