一种面向软件缺陷预测的特征聚类选择方法
    点此下载全文
引用本文:李丽媛,江国华.一种面向软件缺陷预测的特征聚类选择方法[J].计算技术与自动化,2018,(2):126-131
摘要点击次数: 135
全文下载次数: 
作者单位
李丽媛,江国华 (南京航空航天大学 计算机科学与技术学院江苏 南京 210016) 
中文摘要:软件缺陷预测技术通过分析软件静态信息,对软件模块的缺陷倾向性做出判断,合理分配测试资源。但有时搜集的大量度量元信息是无关或冗余的,这些高维的特征增加了缺陷预测的复杂性。文章提出了一种新的度量元选择方法,首先通过样本聚类将相似度高的样本聚在同一簇中,然后在每个簇中按照最低冗余度进行特征子集的挑选,主要选择相互间冗余度低,且预测能力强的度量元。最后通过NASA数据集的实例证明本文方法能有效降低特征子集的冗余率,并能有效提高预测的准确度。
中文关键词:软件缺陷预测  特征选择  特征聚类
 
A Feature Clustering Algorithms for Software Defect Prediction
Abstract:Software defect prediction technology based on the analysis of static information, to judge the defect tendency of software module, and reasonable distribute of test resources. But sometimes collected a large number of measurement information is irrelevant or redundant, these high-dimensional feature increased the complexity of the defect prediction.This paper proposes a new method of feature choice, first of all, through clustering High similarity of the samples in a same cluster,in the same cluster, then ,select feature subset according to the minimum redundancy in each cluster, we will choice the features which have low redundancy, and strong predict ability. Finally through the NASA data sets prove that this method can effectively reduce the redundancy rate of feature subsets, and can effectively improve the accuracy of prediction.
keywords:software defect prediction  feature selection  feature cluster
查看全文   查看/发表评论   下载pdf阅读器