文献类型：期刊文献

中文题名：一种多源异构软件缺陷预测方法

英文题名：Multi-source Heterogeneous Software Defect Prediction Method

作者：杨杰[1,2];范贵生[1];虞慧群[1]

机构：[1]华东理工大学计算机科学与工程系,上海200237;[2]上海市计算机软件测评重点实验室,上海201112

年份：2019

卷号：40

期号：4

起止页码：851

中文期刊名：小型微型计算机系统

外文期刊名：Journal of Chinese Computer Systems

收录：CSTPCD;;Scopus;北大核心:【北大核心2017】；CSCD:【CSCD_E2019_2020】；

基金：国家自然科学基金项目(61702334;61772200)资助;上海市浦江人才计划项目(17PJ1401900)资助;上海自然科学基金项目(17ZR1406900;17ZR1429700)资助;华东理工大学教育教学规律与方法研究项目(ZH1726108)资助

语种：中文

中文关键词：软件缺陷预测;异构度量元;集成学习;相似度

外文关键词：software defect prediction;heterogeneous metrics;ensemble learning;similarity

摘要：现有的异构软件缺陷预测方法解决了跨项目软件缺陷预测中要求源项目与目标项目具有相同度量元的问题,其以度量元相似度为依据匹配源项目与目标项目的度量元,并基于此构建预测模型.然而该方法在计算度量元相似度时存在一定的缺陷且预测性能不稳定.为解决上述问题,提出一种基于度量元相似度的多源异构软件缺陷预测方法,该方法使用数据统计特征计算度量元相似度,并将不同源项目构建的预测模型输出结果根据其相似度大小进行加权整合以得到最终的预测结果.在9组异构项目总计38个数据集上的实验结果表明,该方法显著提高了异构缺陷预测的性能,缩小了与同项目缺陷预测的差距.
The existing heterogeneous software defect prediction method solves the problem that cross-project software defect prediction requires source project must have the same metrics as target project. It matches metrics between source project and target project based on metric similarity and constructs the prediction model by them. However,this method has some defects in calculating metric similarity and its performance is unstable. In order to solve these problems,a multi-source heterogeneous software defect prediction method based on metric similarity is proposed. It calculates the metric similarity by data statistical features and obtains the final prediction result through weighting the outputs of models generated by different source projects according to their similarities. The experimental results on 38 data sets from 10 groups of heterogeneous projects show that this method significantly improves the performance of heterogeneous defect prediction and narrow s the gap with with-in project defect prediction.