基于动态分布适应网络的跨项目缺陷预测

Cross-Project Defect Prediction Based on Dynamic Distributed Adaptive Networks

在软件缺陷预测中,跨项目缺陷预测是基于源项目的标记数据来训练模型,并预测当前正在开发的目标项目的缺陷;然而,两个不同项目数据之间的分布差异往往限制了跨项目缺陷预测模型的能力;由于源域和目标域的数据通常来自不同的分布,因此现有方法主要集中于适应跨域边缘或条件分布;在实际应用中,现有方法无法定量评估边缘分布和条件分布的重要性,这将导致传输性能不理想;论文提出了一种基于动态分布适应网络的跨项目缺陷预测方法来解决分布差异问题,它利用迁移学习能够定量评估每个分布的相对重要性;论文对来自3个公共数据集的24个项目进行了实验,以验证所提出的方法;结果表明,平均而言在AUC和F_(1)分数上分别比所有基线方法高出至少1.3%和5.7%;这表明所提出的方法具有良好的性能特点。

In software defect prediction,cross-project defect prediction is based on training models by using the labeled data from the original project,and predicts the defects of the current development target project.However,the data distribution differences between two different projects often limit the ability of cross-project defect prediction models.As the data from the source domain and target domain usually come from different distributions,existing methods mainly adapt to cross-domain edges or conditional distributions.In practical application,existing methods are unable to quantitatively evaluate the importance of marginal and conditional distributions,which leads to unsatisfactory transmission performance.This paper proposes a cross-project defect prediction method based on a dynamic distribution adaptation network to address the distribution difference,the transmission learning is used to quantitatively evaluate the relative importance of each distribution.the proposed method is verified by the experiments on 24 projects from 3 public datasets.The results show that,on average,the proposed method is superior to all baseline methods,the proposed method reaches the AUC and F_(1) scores by at least 1.3% and 5.7%,respectively.This indicates that the proposed method has good performance characteristics.