结合随机属性与集成的软件缺陷预测算法  

Software defect prediction algorithm combing random attribute and integration

作　　者：王诗博 李勇[1,2] 米文博[1] WANG Shibo;LI Yong;MI Wenbo(College of Computer Science and Technology,Xinjiang Normal University,Urumqi 830054,China;MIIT Key Laboratory of Software Development and Verification Technology for High Security Systems,Nanjing University of Aeronautics and Astronautics,Nanjing 211106,China)

机构地区：[1]新疆师范大学计算机科学技术学院,新疆乌鲁木齐830054 [2]南京航空航天大学高安全系统的软件开发与验证技术工业和信息化部重点实验室,江苏南京211106

出　　处：《现代电子技术》2021年第22期91-96,共6页Modern Electronics Technique

基　　金：国家自然科学基金(61562087);国家自然科学基金(U1703261);新疆自治区高校科研计划项目(XJEDU2017S031);高安全系统的软件开发与验证技术工业和信息化部重点实验室开放基金

摘　　要：软件缺陷预测是提高软件测试效率和保证软件可靠性的重要途径。为了提高软件缺陷预测模型的性能,文中提出一种集成随机属性的朴素贝叶斯算法。该算法通过随机属性子集与若干个朴素贝叶斯基分类器结合后进行集成,实现对软件数据的缺陷预测。首先,构造随机属性子集并将随机属性子集与基模型进行匹配并训练;其次,使用验证集对基模型逐一验证,删除正确率低于随机概率的基模型;最后,将正确率大于随机概率的基模型进行集成,得到最终分类模型。通过使用NASA软件缺陷预测公开数据集进行实验,并与5种常用软件缺陷预测算法进行对比,结果表明,该算法预测率与误报率保持较优的情况下综合评价指标AUC提升0.126。因此,该算法在预测率和误报率保持相对稳定的同时,对缺陷模块的分类效果有大幅提升,具有一定实用价值。The software defect prediction is an important way to improve software testing efficiency and ensure software reliability.A naive Bayesian algorithm integrating random attribute is proposed to improve the performance of software defect prediction model.In the algorithm,the random attribute subsets are combined with several naive Bayesian classifiers to realize the defect prediction for the software data.The random attribute subset is constructed,and the random attribute subset and the base model are matched and trained.The base models are verified one by one by using the verification set,and the base models whose accuracy rate is lower than the random probability are deleted.The base models whose accuracy is greater than the random probability are integrated to get the final classification model.The algorithm are verified with the open data set of NASA software defect prediction and compared with 5 common software defect prediction algorithms.The results show that the comprehensive evaluation index AUC of the algorithm can increase by 0.126 when the prediction rate and false alarm rate of the algorithm remain better.Therefore,the algorithm not only ensures that the prediction rate and false alarm rate remain relatively stable,but also greatly improves the classification effect of defect modules.

关 键 词：软件缺陷预测 随机属性子集 模型匹配 基模型集成 朴素贝叶斯算法 分类模型 

分 类 号：TN806-34[电子电信—信息与通信工程] TP311[自动化与计算机技术—计算机软件与理论]