On Applicability of Cross-project Defect Prediction Method for Multi-Versions Projects
On Applicability of Cross-project Defect Prediction Method for Multi-Versions Projects
复制标题
DOI：
10.1145/3127005.3127015
复制
发表时间：
2017-11
期刊：
影响因子：
--
通讯作者：
S. Amasaki
中科院分区：
文献类型：
--
作者：
S. Amasaki
背景：跨项目缺陷预测（CPDP）研究已受到广泛关注，到目前为止已经提出了许多CPDP方法。由于直接使用跨项目（CP）数据是无效的，这些方法会对目标项目数据对CP数据进行筛选、加权和调整。这个想法对于拥有过往缺陷数据的项目也是有用的。
目的：评估CPDP方法对多版本项目的适用性。评估重点关注性能变化与旧版本发布数据与目标项目的接近程度之间的关系。
方法：我们进行了实验，比较了使用有和没有最近邻（NN）过滤器（一种经典的CPDP方法）的旧版本数据之间的预测性能。
结果：NN过滤器在预测性能方面没有产生明显差异。
结论：NN过滤器对于利用旧版本发布数据提高预测性能没有帮助。
Context: Cross-project defect prediction (CPDP) research has been popular, and many CPDP methods have been proposed so far. As the straightforward use of Cross-project (CP) data was useless, those methods filter, weigh, and adapt CP data for a target project data. This idea would also be useful for a project having past defect data. Objective: To evaluate the applicability of CPDP methods for multi-versions projects. The evaluation focused on the relationship between the performance change and the proximity of older release data to a target project. Method: We conducted experiments that compared the predictive performance between using older version data with and without Nearest Neighbor (NN) filter, a classic CPDP method. Results: NN-filter could not make clear differences in predictive performance. Conclusions: NN-filter was not helpful for improving predictive performance with older release data.