An exploratory study on applicability of cross project defect prediction approaches to cross-company effort estimation
An exploratory study on applicability of cross project defect prediction approaches to cross-company effort estimation
复制标题
DOI：
10.1145/3416508.3417118
复制
发表时间：
2020-11
期刊：
影响因子：
--
通讯作者：
S. Amasaki;Hirohisa Aman;Tomoyuki Yokogawa
中科院分区：
文献类型：
--
作者：
S. Amasaki;Hirohisa Aman;Tomoyuki Yokogawa
背景：软件工作量估计的研究已经活跃了几十年，特别是在开发工作量估计模型方面。工作量估计模型需要从与要估计的项目类似的已完成项目中收集的数据集。这种相似性受到数据集转移的影响，跨公司软件工作量估计（CCSEE）成为一个有吸引力的研究主题。最近一项关于数据集转移问题的研究检验了跨项目缺陷预测（CPDP）方法的适用性和有效性。由于所检查的方法数量有限，不足以得出结论。目的：研究 CPDP 方法的特征，这些方法适用于并有效地解决工作量估计中的数据集转移问题。方法：我们首先回顾了 24 种 CPDP 方法的特点，以寻找适用的方法。接下来，我们使用十个数据集配置研究了它们在工作量估计性能方面的有效性。结果：在 CrossPare 框架中实施的 24 种 CPDP 方法中，有 16 种被发现适用于 CCSEE。然而，只有一种方法可以提高工作量估计性能。大多数其他物质会降解它并且是有害的。结论：我们研究的大多数 CPDP 方法对于 CCSEE 来说都是无能为力的。
BACKGROUND: Research on software effort estimation has been active for decades, especially in developing effort estimation models. Effort estimation models need a dataset collected from completed projects similar to a project to be estimated. The similarity suffers from dataset shift, and cross-company software effort estimation (CCSEE) gets an attractive research topic. A recent study on the dataset shift problem examined the applicability and the effectiveness of cross-project defect prediction (CPDP) approaches. It was insufficient to bring a conclusion due to a limited number of examined approaches. AIMS: To investigate the characteristics of CPDP approaches that are applicable and effective for dataset shift problem in effort estimation. METHOD: We first reviewed the characteristics of 24 CPDP approaches to find applicable approaches. Next, we investigated their effectiveness in effort estimation performance with ten dataset configurations. RESULTS: 16 out of 24 CPDP approaches implemented in CrossPare framework were found to be applicable to CCSEE. However, only one approach could improve the effort estimation performance. Most of the others degraded it and were harmful. CONCLUSIONS: Most of the CPDP approaches we examined were helpless for CCSEE.