面向融合度量的PSO-RBF软件缺陷数量预测模型  

PSO-RBF Software Defect Quantity Prediction Model for Fusion Metrics

作　　者：刘路瑶 韩培胜[1] 李伟群 李万鹏 LIU Luyao;HAN Peisheng;LI Weiqun;LI Wanpeng(Information Engineering University,Zhengzhou 450001,China)

机构地区：[1]信息工程大学,河南郑州450001

出　　处：《信息工程大学学报》2023年第6期691-698,共8页Journal of Information Engineering University

基　　金：国家自然科学基金资助项目(61572517)。

摘　　要：现有软件缺陷预测技术主要基于软件产品度量或过程度量来预测软件缺陷倾向性或缺陷数量,缺乏针对融合产品度量指标和过程度量指标的缺陷预测研究。为提高面向融合度量指标的软件缺陷预测模型的适用性和准确度,以融合软件产品度量指标和过程度量指标为输入,提出一种软件缺陷数量预测模型。此模型主要包括特征选择和缺陷数量预测两阶段。特征选择阶段,采用一种改进密度峰值聚类算法和皮尔逊相关系数结合的特征选择方法,完成特征选取;缺陷数量预测阶段,基于径向基函数(Radial Basis Function, RBF)神经网络,引入粒子群优化(Particle Swarm Optimizer, PSO)算法构建PSO-RBF软件缺陷数量预测模型。实验结果表明,面向融合度量指标的PSO-RBF模型在缺陷数量预测中效果更优。The existing software defect prediction technology mainly predicts the tendency or number of software defects based on software product metrics or process metrics,and there is a lack of defect prediction research on integrating product metrics and process metrics.To improve the applicability and accuracy of the software defect prediction model for fusion metrics,a software defect quantity prediction model is proposed based on the fusion of software product metrics and process metrics.This model mainly includes two stages:feature selection and defect quantity prediction.At the fea-ture selection stage,a feature selection method combining improved density peak clustering algo-rithm and Pearson correlation coefficient is proposed to complete the feature selection;at the defect quantity prediction stage,based on radial basis function(RBF)neural network,particle swarm opti-mization(PSO)algorithm is introduced to build PSO-RBF software defect quantity prediction model.The experimental results show that the PSO-RBF model for fusion metrics is more effective in defect quantity prediction.

关 键 词：软件缺陷数量预测 融合度量指标 密度峰值聚类算法 粒子群优化算法 径向基函数神经网络 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论]