摘要
为提高无标识软件缺陷预测的准确性，提出一种谱聚类与混沌免疫相结合的软件缺陷预测方法。该方法首先将谱聚类算法引入到软件缺陷预测领域中，然后针对谱聚类算法中K-Means局部收敛的缺点，用一种混沌免疫聚类算法来替换K-Means算法。同时，在免疫克隆选择算法的框架下，借鉴混沌和免疫理论，设计免疫克隆聚类适应度函数计算方法，并给出分层混沌变异算子，以实现种群多样性的增加，促进无标识软件缺陷数据预测精度的提高。在Iris和3组商业软件模块数据集上进行了仿真实验，实验结果验证了该方法的有效性。
To improve the accuracy of defect prediction for unlabeled software data sets, a novel sottware aelect preo^cuo,, method based on the combination of spectral clustering and chaotic immune is presented. The method first intro- duces the Ng-Jordan-Weiss （NJW） algorithm, a spectral clustering algorithm, into the field of software defect pre- diction, and then uses a new chaotic immune clustering algorithm go replace the K-Means algorithm to overcome the K-Means＇ s problem of easily getting trap local optima in spectral clustering. And under the framework of immune clone selection, it designs a new affinity function for immune clone clustering and gives the layered chaotic mutation operator based on the immune and chaotic theory to diversify the antibodies and improve the accuracy of software defect prediction. Two ease studies are used to validate the method on the Iris and three commercial software data sets. The experimental results illustrate the effectiveness of the proposed method.
出处
《高技术通讯》
CAS
CSCD
北大核心
2012年第12期1219-1224,共6页
Chinese High Technology Letters
基金
863计划（2010AA7010213）,国家自然科学基金（61179005,61179004）和十一五国防预研（513270104）资助项目.
关键词
无标识数据
免疫
谱聚类
混沌
软件缺陷预测
unlabeled data, immune, spectral clustering, chaos, software defect prediction
