基于模糊聚类非负矩阵分解的软件缺陷预测  

Applying Non-Negative Matrix Factorization Based on Fuzzy C-Means to Software Defect Prediction

作　　者：常瑞花[1] 慕晓冬[1] 李琳琳[1] 宋国军[1] 

机构地区：[1]第二炮兵工程学院,西安710025

出　　处：《宇航学报》2011年第9期2059-2064,共6页Journal of Astronautics

基　　金：国防重点预研项目(513270104);国家高技术研究发展项目(2010AA7010213)

摘　　要：为了进一步提高对不平衡软件缺陷数据的预测精度,提出了一种基于模糊聚类非负矩阵分解的缺陷预测方法。该方法首先选取最近邻算法作为不平衡缺陷数据预测的基分类器,然后将非负矩阵分解算法引入软件缺陷预测领域,并提出利用模糊聚类初始化非负矩阵分解算法的思路。该方法一方面解决了最近邻算法处理多属性缺陷数据时计算量大导致性能降低的问题,另一方面克服了非负矩阵分解的随机初始化导致结果陷入局部最优的不足,提高了不平衡软件缺陷数据的预测精度。最后通过两组算例验证了方法的有效性。In order to improve the prediction accuracy for imbalanced software defect data,a novel method is presented.Firstly,a K-Nearest Neighbour（KNN） algorithm is chosen as a basic classifier.Then Non-negative Matrix Factorization（NMF） algorithm is introduced to mine the imbalanced software defect data for potential characters.Its purpose is to reduce the computational effort while the KNN algorithm processing lots of attributes.Meantime,aim at the problem of NMF easily falling into local-optimal resulted from random initialization,a Fuzzy C-means（FCM） method is used to initialize its basic matrix,thus improving the prediction accuracy.Finally,two case studies are used to validate the method.The experimental results show that the proposed method is effective.

关 键 词：软件缺陷 非负矩阵分解 模糊聚类 预测 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论]