Heterogeneous defect prediction with two-stage ensemble learning
Heterogeneous defect prediction with two-stage ensemble learning
复制标题
通过两阶段集成学习进行异构缺陷预测
DOI：
10.1007/s10515-019-00259-1
复制
发表时间：
2019
影响因子：
3.4
通讯作者：
Ying Shi
中科院分区：
文献类型：
--
作者：
Li Zhiqiang;Jing Xiao-Yuan;Zhu Xiaoke;Zhang Hongyu;Xu Baowen;Ying Shi
异构缺陷预测（HDP）是指使用从其他项目（源）收集的异构数据来预测一个项目（目标）中容易出现缺陷的软件模块。近年来，人们提出了几种HDP方法。然而，这些方法没有充分结合缺陷数据的两个特征：(1)数据可能是线性不可分割的，(2)数据可能是高度不平衡的。这两个数据特征使得建立一个有效的HDP模型具有挑战性。本文提出了一种新的两阶段集成学习（TSEL）方法，该方法包括两个阶段：集成多核域自适应（EMDA）阶段和集成数据采样（EDS）阶段。在EMDA阶段，我们开发了一个集成多核相关比对（EMKCA）预测器，它结合了多核学习和领域自适应技术的优势。在EDS阶段，我们采用RESample with replacement （RES）技术来学习多个不同的EMKCA预测因子，并使用平均集成将它们组合在一起。这两个阶段创建了缺陷预测器的集合。在30个公共项目上进行的大量实验表明，所提出的TSEL方法优于一系列竞争方法。auc改善20.14 ~ 33.92%，测量改善36.05 ~ 54.78%，不平衡改善5.48 ~ 19.93%。
Heterogeneous defect prediction (HDP) refers to predicting defect-prone software modules in one project (target) using heterogeneous data collected from other projects (source). Recently, several HDP methods have been proposed. However, these methods do not sufficiently incorporate the two characteristics of the defect data: (1) data could be linear inseparable, and (2) data could be highly imbalanced. These two data characteristics make it challenging to build an effective HDP model. In this paper, we propose a novel Two-Stage Ensemble Learning (TSEL) approach to HDP, which contains two stages: ensemble multi-kernel domain adaptation (EMDA) stage and ensemble data sampling (EDS) stage. In the EMDA stage, we develop an Ensemble Multiple Kernel Correlation Alignment (EMKCA) predictor, which combines the advantage of multiple kernel learning and domain adaptation techniques. In the EDS stage, we employ RESample with replacement (RES) technique to learn multiple different EMKCA predictors and use average ensemble to combine them together. These two stages create an ensemble of defect predictors. Extensive experiments on 30 public projects show that the proposed TSEL approach outperforms a range of competing methods. The improvement is 20.14–33.92% inAUC, 36.05–54.78% inf-measure, and 5.48–19.93% inbalance, respectively.