面向演化的缺陷预测模型演进策略比较研究  

Comparative Study of Evolution Oriented Defect Prediction Model Evolution Strategies

作　　者：刘文广[1,2] 高东静[1,2] 彭鑫[1,2] 赵文耘[1,2] 

机构地区：[1]复旦大学软件学院,上海201203 [2]上海市数据科学重点实验室(复旦大学),上海201203

出　　处：《小型微型计算机系统》2017年第4期671-677,共7页Journal of Chinese Computer Systems

基　　金：国家自然科学基金项目(61370079)资助

摘　　要：目前,已有许多学者对缺陷预测进行了相关的研究.但这些研究主要关注单次预测.然而软件项目是持续演化的;缺陷分布规律也在持续变化.因此,缺陷预测模型也必须随着软件演化而不断演进.为了分析预测模型演进策略的性能,本文整理了面向软件演化的缺陷预测的基本流程和五种常用的缺陷预测模型演进策略.在此基础上,将五种演进策略应用到6个开源项目的演化数据中,并进行了相关的比较分析.结果表明:1)Boosting缺陷预测模型演进策略有较好的预测性能;2)对于Boosting模型演进策略,何种参数组合具有的最佳预测性能具有项目相关性.Many scholars had studied on defect prediction, and these studies mainly concerned on one-time prediction. However, soft- ware projects, as well as the defect distribution, are often in continuous evolution. Therefore, software defect prediction model must e- volve with software evolution. For this problem, the paper firstly summarizes the software evolution defect prediction process, and five common defect prediction model evolution strategies. Then experiments on 6 open source projects are performed with different model evolution strategies. The experimental results show ： 1 } the boosting-based model evolution strategy has best performance ;2 ） the per- formance of boosting-based model evolution strategy under different parameter is project-specific.

关 键 词：提升方法 引导聚集 软件演化 缺陷预测 模型演进策略 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论]