跨项目软件缺陷预测方法研究综述

Review on Cross-project Software Defects Prediction Methods

软件缺陷预测是提高软件测试效率、保证软件可靠性的重要途径,已经成为目前实证软件工程领域的研究热点。在软件工程中,软件的开发过程或技术平台可能随时变化,特别是遇到新项目启动或旧项目重新开发时,基于目标项目数据的传统软件缺陷预测方法无法满足实践需求。基于迁移学习技术采用其他项目中已经标注的软件数据实现跨项目的缺陷预测,可以有效解决传统方法的不足,引起了国内外研究者的极大关注,并取得了一系列的研究成果。首先总结了跨项目软件缺陷预测中的关键问题。然后根据迁移学习的技术特点将现有方法分为基于软件属性特征迁移和软件模块实例迁移两大类,并分析比较了常见方法的特点和不足。最后探讨了跨项目软件缺陷预测未来的发展方向。

Software defect prediction is an important way to improve the software testing efficiency and ensure software reliability,which has become a research hotspot in the field of empirical software engineering.In software engineering,the software development process or technology platform may change at any time.Especially when a new project is started or an old project is redeveloped,the traditional within-project software defect prediction method cannot meet the practical needs.Cross-project software defect prediction that using the cross-project labeled data and transfer learning technology can effectively solve the shortcomings of traditional method,which has attracted great attention of scholars at home and abroad,and produced a series of research findings.Firstly,the key problems of cross-project software defect prediction methods are summarized.Then,according to the technical characteristics of transfer learning,the existing methods are divided into two types,i.e.,the methods based on attribute characteristics and the methods based on software module instances,and the characteristics and shortcomings of common methods are analyzed and compared.Finally,the future development direction of cross-project software defect prediction is discussed.