基于主动学习的软件缺陷预测实证研究  

An Empirical Study on Software Defect Prediction Based on Active Learning

作　　者：曲豫宾 QU Yubin(Jiangsu College of Engineering and Technology,Nantong 226007,China)

机构地区：[1]江苏工程职业技术学院

出　　处：《江苏工程职业技术学院学报》2019年第2期11-15,共5页Journal of Jiangsu College of Engineering and Technology

基　　金：南通市科技计划指令性项目“即时软件缺陷预测模型构建和优化关键技术研究”(编号JC2018134)

摘　　要：软件缺陷在软件运行过程中可能会产生意料之外的行为,造成重大损失。软件缺陷预测模型需要大量的数据集标注,存在工作量大、成本较高等问题。主动学习可从未标注数据集中选择辨识度高的样例进行人工标注,从而提升软件缺陷预测模型建立的效率。在AEEEM公开数据集上对常见的4种主动查询策略(RANDOM、COMMITTEE、QUIRE、UNCERTAINTY)进行实证研究,结果表明,基于信息熵的不确定度策略UNCERTAINTY在数据集上有优于其他查询策略的表现。Unexpected conditions owing to software defects will arise in the process of software operation and hence cause great losses.The software defect prediction model needs a large number of labeled datasets,which requires heavy work and costs a lot.Through active learning,easily recognizable data samples can be selected from the unlabeled datasets and get labeled manually to improve the efficiency of the software defect prediction model.Four common active learning strategies(RANDOM,COMMITTEE,QUIRE,UNCERTAINTY)are studied on AEEEM open datasets and the results indicate that UNCERTAINTY based on information entropy has better performance than other query strategies.

关 键 词：主动学习 软件缺陷预测 信息熵 查询策略 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]