基于差分演化算法的跨项目缺陷预测方法  

Cross-project defect prediction method based on differential evolution algorithm

作　　者：范贵生[1,2] 陈玎乐 杨星光 虞慧群 FAN Gui-sheng;CHEN Ding-le;YANG Xing-guang;YU Hui-qun(Department of Computer Science and Engineering,East China University of Science and Technology,Shanghai 200237,China;Shanghai Key Laboratory of Computer Software Evaluating and Testing,Shanghai Computer Software Technology Development Center,Shanghai 201112,China)

机构地区：[1]华东理工大学计算机科学与工程系,上海200237 [2]上海计算机软件技术开发中心上海市计算机软件评测重点实验室,上海201112

出　　处：《计算机工程与设计》2022年第7期1890-1897,共8页Computer Engineering and Design

基　　金：国家自然科学基金项目(61702334、61772200);上海市浦江人才资助计划基金项目(17PJ1401900);上海市自然科学基金项目(17ZR1406900、17ZR1429700);华东理工大学教育科研基金项目(ZH1726108);上海应用技术学院资助合作创新基金会基金项目(XTCX2016-20)。

摘　　要：为提高工作量感知的跨项目缺陷预测性能,提出一种基于差分演化算法的建模方法DE-EACPDP。提出缺陷密度百分位数平均(FDPA)作为模型在训练集上的优化目标,采用逻辑回归(LR)建立预测模型;为使LR在训练集上得到最大的FDPA,使用差分演化算法搜索最优的LR系数。实验在4个数据集的82个项目上对模型进行评估,其结果表明,与5个最先进的方法相比,DE-EACPDP可以显著提升工作量感知性能。To improve the performance of effort-aware cross-project defect prediction,a modeling method DE-EACPDP based on differential evolution algorithm was proposed.The fault-density-percentile-average(FDPA)was proposed as the optimization target of the model on the training set,and the logistic regression(LR)was used to establish the prediction model.To make LR get the largest FDPA on the training set,the differential evolution algorithm was used to search for the optimal LR coefficient.The model was evaluated on 82 projects in four datasets.The results show that compared with the 5 most advanced methods,DE-EACPDP can significantly improve the effort-aware performance.

关 键 词：软件缺陷预测 跨项目 工作量感知 差分演化 经验软件工程 

分 类 号：TP311.5[自动化与计算机技术—计算机软件与理论]