2016年南京大学《软件质量保证项目中的数据挖掘系统》(肖韬计算机系).pdfVIP

1000-9825/2004/15(01)0000 ©2004 Journal of Software 软 件 学 报 Vol.15, No.1 软件质量保证项目中的数据挖掘系统 ——基于半监督学习与SVMlight 肖韬 (南京大学 计算机科学与技术系,江苏 南京 210093) A Data Mining System for Software Quality Assurance Project ——Based on Semi-supervised Learning and SVMlight Xiao Tao (Department of Computer Science and Technology, Nanjing University, Nanjing 210093, China) Abstract : The principal purpose of this data mining system is to train a classifier based on given labeled records in relatively small quantity and unlabeled records in relatively large quantity, such that we can deploy intense testing work on suspicious models. To handle the two difficulties in this task, namely the imbalanced data distribution and low percentage of labeled records in the whole data set, methods of data set slicing-and-reconstructing and SVMlight based on semi-supervised learning are adopted. On top of SVMlight based on semi-supervised learning, this paper tried the SVM algorithm which uses only the labeled data, so as to contrast the performan ce of these two classifying algorithms. Finally, ten-fold cross-validation is carried out to test the classifying performance for these algorithms and testing results are given out. Key words: data mining ；classification ；data set slicing-and-reconstructing ；semi-supervised learning ；transductive learning ；imbalanced data distribution ；SVMlight ；polling algorithm ；ten-fold cross-validation 摘 要: 本系统的主要任务是根据给定的少量已标记数据与大量未标记数据训练出一个分类器，预测给定的 软件模块是否含有可能的缺陷，以便对可能的缺陷模块重点展开测试。该任务的主要难点是数据分布的不均 衡性和已标记数据在测试数据集中的比重过低，为此本系统采用了数据集切分重构法来解决第一个难点，采 用基于半监督学习的 SVMlight 来解决第二个难点，并在这二者的基础之上使用投票算法来决定最终的分类结 果。除了采用基于半监督学习的SVMlight 来做分类器之外，本文还尝试了使用仅仅基于已标记数据的SVM 算 作者简介: 肖韬( 1985－),男, 江苏省南京市人, 硕士研究生在读,主要研究领域为计算机体系结构与并行计算. 随米文库，考试资料分享、查找、打印，一站式解决! 2 Journal of Software