统计策略序列模式挖掘及其在软件缺陷预测中的应用  

Statistically Significant Sequential Pattern Mining Applying to Software Defect Prediction

作　　者：唐磊[1] 李春平[1] 杨柳[2] 

机构地区：[1]清华大学软件学院,北京100084 [2]中南大学软件学院,长沙410075

出　　处：《计算机科学》2013年第5期164-167,188,共5页Computer Science

摘　　要：人类的生活越来越依赖于高可靠性和可用性的软件系统,软件缺陷一直是软件工程领域中研究最活跃的内容之一。在研究序列模式挖掘技术的基础上,介绍了软件缺陷预测的相关技术,设计了一种基于统计策略的序列模式挖掘算法的软件缺陷预测方案,实现了InfoMiner和STAMP两种模式挖掘算法、卡方检验特征选择和SVM等分类算法;构造了一个软件缺陷预测模型,实现了预测和发现软件系统中的未知缺陷的功能。实验结果表明,所提软件预测模型可以获得良好的预测结果,具有一定的使用价值和应用前景。Nowadays the human beings are more and more reliant on software systems which have high reliability and usability, and the technology of software defect prediction has been one of the most active parts of software engineering. This paper introduced the technology of software defect prediction on the basis of sequential pattern mining and de- signed a model for software defect prediction with the technology of mining statistically significant pattern. It described the architecture and detailed implementation of the algorithms named ＂InfoMiner＂ and ＂STAMP＂. The model using In- foMiner and STAMP to mine patterns, chi-square test to feature selection and SVM to classify can find unknown defects with high probability. Experimental results show that the model is able to get high prediction accuracy, so that it is valua- ble and has future prospects.

关 键 词：数据挖掘 序列模式 软件缺陷 信息增益 分类预测 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]