摘要
为解决通常由公司内工程数据训练构建软件缺陷预测模型,而实际较缺乏本地缺陷数据的问题,借助迁移学习技术,提出使用不同公司工程数据构建缺陷预测模型的算法,实现跨公司软件缺陷预测。通过比较源工程训练数据和目标工程测试数据集的多种统计量,设置训练数据的权重,基于加权的训练数据构建加权迁移朴素贝叶斯分类器。实验结果表明,该方法有效提高了跨公司软件缺陷预测模型的性能,为项目管理者合理分配软件工程资源提供了依据。
To address the problem that building software defect prediction models usually uses within-company project data that is always scarce in practice,using cross-company project data,a cross-company defect prediction algorithm was proposed with the help of transfer learning technique.The weight of training data from source project was set by comparing multiple statistics of training data with those of target project data set.A weighted transfer Naive Bayes classifier was built based on weighted training data.Experimental results show that the proposed method can effectively improve the performance of cross-company software defect prediction and optimize software resource allocation for project managers.
出处
《计算机工程与设计》
北大核心
2016年第3期684-689,共6页
Computer Engineering and Design
基金
国家自然科学基金项目(61462048)
九江学院科研基金项目(2014KJYB019
2015LGYB26)
关键词
软件缺陷预测
迁移学习
机器学习
朴素贝叶斯
软件度量
software defects prediction
transfer learning
machine learning
Naive Bayes
software metrics