基于流形学习降维的缺陷动态特征分类算法  

Defect Dynamic Feature Classification Algorithm Based on Manifold Learning Dimensionality Reduction

作　　者：汪绍荣 黄卫春[2] 宗波 WANG Shao-rong;HUANG Wei-chun;ZONG Bo(School of Mathematics and Computer Science,Yichun University,Yichun Jiangxi 336000,China;Software College,East China Jiaotong University,Nanchang Jiangxi 330013,China)

机构地区：[1]宜春学院数学与计算机科学学院,江西宜春336000 [2]华东交通大学软件学院,江西南昌330013

出　　处：《计算机仿真》2023年第11期475-479,共5页Computer Simulation

基　　金：江西省教育厅科学技术研究项目(GJJ2210706)

摘　　要：开源软件缺陷报告分类过程中,缺陷报告数据维度不统一会直接降低后续报告的分类精度,为此提出基于行为路径树的开源软件缺陷报告分类算法。使用流形学习降维算法对开源软件缺陷报告数据实施降维处理,有效去除噪声数据,初步提升后续报告分类的分类精度;使用行为路径树对降维报告数据实施缺陷动态特征提取,进一步强化分类精度;根据随机森林原理建立开源软件缺陷报告分类模型,完成开源软件缺陷报告的精准分类。实验结果表明,使用上述方法开展缺陷报告分类时,分类精度高、效果好。In the process of classifying open-source software defect reports,uneven data dimensions may directly reduce the classification accuracy of subsequent reports.Therefore,a classification algorithm for open-source software defect reports based on behavioral path tree was proposed.At first,a manifold learning dimensionality reduction algo-rithm was adopted to reduce the dimensionality of open-source software defect data,thus effectively removing noise data and initially improving the classification accuracy of subsequent reports.Moreover,behavioral path trees were used to extract the defect dynamic features from dimensionality reduction report data and thus to further enhance clas-sification accuracy.After that,a classification model was built for open-source software defect reports based on the random forest principle.Finally,accurate classification for open-source software defect reports was achieved.Experi-mental results show that using this method has higher classification accuracy.

关 键 词：行为路径树 开源软件 缺陷报告 分类算法 特征提取 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论]