改进加权朴素贝叶斯的软件缺陷预测算法

针对软件缺陷预测领域特征之间存在紧密关联性而影响朴素贝叶斯分类性能的问题,提出一种改进加权朴素贝叶斯的分类算法.首先,通过预处理步骤以及特征成对计算,创建彼此之间的依赖关系.然后,通过构造权值的方式实现朴素贝叶斯的独立性假设松弛.最后,使用离散化方法将软件指标的数值转化为分类值,并利用min-max归一化程序对数据进行...

A Software Defect Prediction Algorithm Using Improved Weighted Naive Bayesian

For the issue that the close correlation between features in software defect prediction applications affects the performance of naive Bayesian classification, an improved weighted naive Bayesian classification algorithm is proposed. Firstly, the dependencies between them are created by pre-processing steps and pairwise computation of features. Then, the independence hypothesis of naive Bayesian is relaxed by constructing weights. Finally, the value of software index is transformed into the classification value by discretization method, and the data is normalized by min-max normalization program. The experiments are carried out by using widely recognized NASA promise data sets. The results show that the proposed algorithm has better prediction effect than the standard naive Bayesian method, and it is more competitive than other feature-weighted techniques.