摘要
研究了在区分故障严重程度下的软件模块故障倾向预测方法,将故障分为高严重程度和低严重程度两种类型,用统计分析和机器学习方法分析静态代码度量与故障倾向之间的关系。以公开和私有两种类型的失效数据集作为实验数据,分析发现,故障的严重程度影响预测性能,预测不同严重程度的故障需要选择不同的度量和分类模型,预测低严重程度故障的性能好于预测高严重程度故障的性能。
Utilizing a public dataset and a private dataset,we employed the statistics analysis and machine learning methods to empirically investigate the relation between static code metrics and the proneness of software modules.We conclude that fault severity impacts the performance of fault-proneness prediction,and we should take the fault severity into account when choosing appropriate metrics and classification models.We also conclude that the performance of prediction for low severity faults is better than the high severity faults.
出处
《武汉大学学报(信息科学版)》
EI
CSCD
北大核心
2010年第5期562-565,共4页
Geomatics and Information Science of Wuhan University
基金
国防预研基金资助项目(513270104)
关键词
软件模块故障倾向
故障严重程度
统计分析
机器学习
fault proneness of software modules
fault severity
statistics analysis
machine learning