一种新型软件缺陷预测模型研究  

New prediction model for software defect-proneness

作　　者：韩璐[1,2] 荆晓远[2] 

机构地区：[1]南京邮电大学计算机学院,江苏南京210023 [2]南京邮电大学自动化学院,江苏南京210023

出　　处：《南京邮电大学学报（自然科学版）》2015年第1期95-101,共7页Journal of Nanjing University of Posts and Telecommunications：Natural Science Edition

基　　金：江苏省研究生培养创新工程(CXLX11_0418);江苏省自然科学基金青年(BK20140888)资助项目

摘　　要：对软件度量元的选择问题是软件缺陷预测中的重要研究内容之一。文中通过采用互信息(MI)来计算度量元与度量元、度量元与目标类别之间的相关性,提出了信息损失最小准则,并将该准则作为Filter模型中消除冗余度量元的依据,然后结合现有的Wrapper模型来选择最终对分类有帮助的度量元集合。该方法既能减轻Wrapper模型高计算成本的负担,又能弥补单纯使用Filter模型对软件缺陷的预测能力相对较弱的缺点。实验表明:该方法在召回率、F-measure值上都有明显的提高,具有一定的有效性和实用性。Software metric selection problem is one of the important research topics of software defectproneness prediction. In this study,the relevance among different software metrics,and the software metrics and the decision of the software modules is measured by mutual information（ MI）,followed by a minimum information loss criterion. Furthermore,based on the criterion,the redundant metrics are removed in the Filter model. And by combining the Wrapper model,some certain metrics that are helpful for the final classification are selected. The proposed method can lessen the high computational burden of Wrapper model and obtain better prediction performance than Filter model. The simulation and the analysis show that the method has significant improvenent in recall rate and F-measure,and thus the method is effective and efficient for software defect prediction.

关 键 词：软件度量元选择 FILTER WRAPPER 软件缺陷预测 信息损失最小 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]