改进的采样算法与无监督聚类相结合的软件缺陷预测模型  

The Software Defect Prediction Model Combining Improved Sampling Algorithm and Unsupervised Clustering

作　　者：石海鹤[1] 周世文 钟林辉[1] 肖正兴 SHI Haihe;ZHOU Shiwen;ZHONG Linhui;XIAO Zhengxing(School of Computer and Information Engineering,Jiangxi Normal University,Nanchang Jiangxi 330022,China;School of Artificial Intelligence,Shenzhen Polytechnic University,Shenzhen Guangdong 518055,China)

机构地区：[1]江西师范大学计算机信息工程学院,江西南昌330022 [2]深圳职业技术大学人工智能学院,广东深圳518055

出　　处：《江西师范大学学报（自然科学版）》2024年第3期301-310,共10页Journal of Jiangxi Normal University(Natural Science Edition)

基　　金：国家自然科学基金(62062039,61872123);教育部高等学校科学研究发展中心专项课题(ZJXF2022255);江西师范大学研究生创新基金(YJS2022027)资助项目。

摘　　要：该文首先在自适应综合过采样算法ADASYN(adaptive synthetic sampling)的基础上,考虑少数类内部不同密度簇之间的连接性问题,将与采样点距离为中等的点纳入新样本生成范围,改进得到T-ADASYN过采样优化算法,有效地增加了少数类内部不同密度簇的连接性,生成了分布更为均衡的数据集.然后使用基于连接的spectral clustering算法进行聚类预测操作,将过采样算法和无监督聚类相结合,提出一种新型实用的软件缺陷预测模型TA-SC(T-ADASYN+spectral clustering).以F-score为评价指标,spectral clustering为聚类模型进行验证.实验结果表明:改进的T-ADASYN过采样算法在公开的PROMISE数据集和NASA数据集上比常用的过采样算法均有6%的性能提升,且TA-SC模型在PROMISE和NASA 2个数据集上比常用聚类算法分别有3%和2%的性能提升.Firstly,based on adaptive comprehensive oversampling algorithm ADASYN(adaptive synthetic sampling),considering the connectivity among different density clusters within a small number of classes,the points that are middle neighbors distance from sampling points are included in the range of new samples,and the T-ADASYN oversampling optimization algorithm is obtained.The T-ADASYN oversampling optimization algorithm is improved to effectively increase the connectivity of clusters with different densities within a few classes and generate a more balanced data set.The connectivity-based Spectral Clustering algorithm is further used for the clustering prediction operation,thus combining the oversampling algorithm and unsupervised clustering for the first time and proposing a novel and practical software defect prediction model TA-SC(T-ADASYN+Spectral Clustering).Using F-Score as the evaluation indicator and Spectral Clustering as the clustering model for validation,the experimental results show that the improved T-ADASYN oversampling algorithm has an average improvement of 6%and 6%compared to commonly used oversampling algorithms on the publicly available PROMISE dataset and NASA dataset,respectively,and the TA-SC model has the highest results of 3%and 2%improvement compared to commonly used clustering algorithms in both datasets.

关 键 词：软件缺陷预测 类别不平衡 过采样算法 聚类算法 无监督学习 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论]