基于演化数据的软件缺陷预测性能改进

软件持续演化已经是不争的事实，演化意味着需求的变化，也就必然导致了缺陷的不断产生.现有的缺陷预测技术多偏重于基于软件工作制品，如文档、代码、测试用例等的属性来预测缺陷，但如果把软件看作一种物种，其生命周期内的演化本质上是一个物种的逐步进化，其缺陷的表现也必然带着该物种的特征，而且还受到进化历史中的演化轨迹的影响.已有一些研究人员开始研究软件演化过程，并提出了一些演化度量元.研究和提出了可以刻画软件演化轨迹的两类演化度量元，并通过案例研究，建立缺陷预测模型.在6个著名开源软件数据集上训练和验证了由软件演化度量元建立的缺陷预测模型，获得了良好的预测性能，验证了演化度量元对缺陷预测性能的改进.

Improving the Performance of Defect Prediction Based on Evolution Data

It is an undisputed fact that software continues to evolve. Software evolution is caused by requirement changes which often result in injection of defects. Existing defect prediction techniques mainly focus on utilizing the attributes of software work products, such as documents, source codes and test cases, to predict defects. Consider an evolving software as a species and its development process as a natural species'' evolutionary process, the injection of defects may have the characters of a species and will be impacted by its evolution. A great many of researchers have studied the process of software evolution and proposed some evolution related metrics. In this study, a set of new metrics is first proposed based on evolutionary history to characterize software evolution process, and then a case study on building defect prediction models is presented. Experiments on six well-known open source projects achieved good performance, demonstrating the effectiveness of the proposed metrics.