基于机器学习的虚拟仪器软件缺陷预测模型研究  

Research on software defect prediction model of virtual instrument based on machine learning

作　　者：曾路 汪浩 ZENG Lu;WANG Hao(Uizhou Power Grid Corporation Minisitry of Science and Technology Information,Guizhou 550002,China)

机构地区：[1]贵州电网有限责任公司信息中心,贵阳550003

出　　处：《自动化与仪器仪表》2020年第5期59-62,共4页Automation & Instrumentation

基　　金：贵州省重大专项(No.黔科合JZ字〔2014〕2001)。

摘　　要：针对传统虚拟仪器软件预测模型易产生偏向无缺陷类别结果,导致出现缺陷预测性能较低的问题,提出基于机器学习的虚拟仪器软件缺陷预测模型。利用邻域清除算法进行数据预处理,消除重叠非缺陷样本,通过随机采样的方式,划分平衡训练集,使用朴素贝叶斯NB算法将训练集映射给定测试样本集中进行软件缺陷预测,得到多个缺陷预测子集,利用机器学习集成得到最终软件缺陷预测模型,完成基于机器学习的虚拟仪器软件缺陷预测模型构建。对比实验结果显示,设计软件缺陷预测模型在不同标记比例的9个程度不同不平衡数据集上,平均赢得的数据集数目为7.67,AUC值均较传统缺陷预测模型有较明显地提高,表明设计软件缺陷预测模型在不同软件数据集上有良好预测效果,为实际虚拟仪器软件缺陷预测提供了新思路。Aiming at the problem that traditional virtual instrument software prediction model is prone to produce defect free category results,which leads to low defect prediction performance,a virtual instrument software defect prediction model based on machine learning is proposed.The neighborhood sweeping algorithm is used for data preprocessing to eliminate the defect samples of overlapping fees,and the equilibrium training set is divided by means of random sampling.The naive bayes NB algorithm is used to map the training set to a given test sample set for software defect prediction,and multiple defect prediction subsets are obtained.The final software defect prediction model is obtained by machine learning integration,and the construction of virtual instrument software defect prediction model based on machine learning is completed.Contrast experiment results show that the design of software defect prediction model at different levels in different markup percentage of nine unbalanced data sets,the average winning the number of data sets is 7.67,AUC values have more apparent than that of traditional defect prediction model,that the software defect prediction model in different design software data sets have good prediction effect,to the actual virtual instrument software defect prediction provides a new way of thinking.

关 键 词：机器学习 虚拟仪器软件 软件缺陷预测 预测模型 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论]