一种基于特征置换的朴素贝叶斯分类器

朴素贝叶斯分类方法是一种广泛使用的分类算法,在独立性假设不完全满足的情况下计算效率和分类效果均较为理想.通过分析全局特征向量中各特征与类别属性之间的联系,提出将组合特征置换多源特征,用组合特征的共现率对多源特征进行概率调整的新方法,在不同数据集的实验中,调整后的朴素贝叶斯分类器(FRNB)的分类精度均好于传统朴素贝叶斯分类器.测试结果表明,改进后的算法是有效可行的.

A na(i)ve Bayesian classifier based on feature replacement

Nave Bayesian classifier is a widespread classification algorithm,in which both the calculation efficiency and classifying effect are comparatively ideal even if the independence supposition is not completely satisfied.By analyzing the connection of every features with class attributes of the global characteristic vector,new method was presented,in which the multi-source feature was replaced with the combined feature and the coexistence rate of the combined feature was used for probability adjustment of the multi-source feature.In the experiment on different data set,the precision of adjusted nave Bayesian classifier(FRNB) was better than that of the traditional nave Bayesian classifier.The result of test indicated that this improved algorithm was effective and feasible.