基于特征加权的多关系朴素贝叶斯分类模型

Multi-relational Nave Bayesian Classifier Using Feature Weighting

为进一步提高多关系朴素贝叶斯方法的分类准确率,分析了已有的特征加权方法,并在将特征加权方法扩展到多关系的情况下结合元组ID传播方法和面向元组的统计计数方法,建立了基于特征加权的多关系朴素贝叶斯分类模型(MRNBC-W)。标准数据集上的实验结果显示,新方法可以在不增加算法时间复杂度的前提下,有效提高金融数据集的分类准确率。文中也给出了结合扩展互信息标准对属性进行过滤后,加权方法和不加权方法的分类比较。

To improve the accuracy of multi-relational naive Bayesian classifiers, this paper discussed existing feature weighting methods and upgraded the method to deal with multi-relational data directly. Based on the tuple ID propagation method and counting methods towards tuples, a multi-relational naive Bayesian classifier using feature weighting （MRNBC-W） was given. Experiments on Financial database show that with the help of feature weighting, the classifiers can give better accuracy without increase of time complexity. Furthermore, MRNBC-W based on mutual information （MRNBC-W-MI） was implemented.