基于聚类分析与Box-Cox转换的跨项目缺陷预测方法

Box-Cox transformation based ensemble learning approach for cross-project software defect prediction

软件缺陷预测需要通过一定的方法预先识别出项目内潜在的程序缺陷,提高软件产品的整体质量。文章深入探究了跨项目缺陷预测问题,对训练集的选择使用聚类分析的方法。在对聚类分析方法使用之前,使用Box-Cox转换来提高聚类分析的性能,由此比较使用度量元转换构建的跨项目缺陷预测模型与未使用度量元转换的预测模型的性能。文章基于实际的数据集,验证了Box-Cox转换的有效性,使用了多种分类器来构建缺陷模型,保证实验结果的广泛性。

Software defect prediction needs to identify potential program defects in the project in advance through certain methods, so as to improve the overall quality of software products. In this paper, the problem of cross-project defect prediction is deeply explored. Cluster analysis is used to select training set. Before using clustering analysis method, a Box-Cox transformation is used to improve the performance of clustering analysis, and the performance of cross-project defect prediction model constructed by metric transformation is compared with that of prediction model without metric transformation. Based on the actual data set, this paper verifies the effectiveness of the Box-Cox transformation. A variety of classifiers are used to construct the defect model to ensure the universality of the experimental results.