基于关联规则的ABN分类器构造

Naive Bayes分类建立在贝叶斯理论基础上，应用极为广泛，它采用类条件独立假设对贝叶斯理论进行了近似。Bayesian Network则在这一基础上采用图形模型弥补了独立假设的不足，同时揭示出分类过程中会导致NP问题的出现。本文采用一种折衷的方法--联合关联规则与ABN分类技术构造贝叶斯分类器。它弥补了独立假设的不足，同时也避免了解决NP问题。最后，本文用实验结果展示它在多个领域远远优于Naive Bayes分类器。

The Structure of ABN Classifiers Based on Association Rules

The Naive Bayesian classification based on the Bayesian theorem is very popular. It approximates the Bayesian theorem by the assumption of class conditional independence. On this basis , Bayesian networks compensate for the assumption by adopting graphical models. On the other hand, it implies that the NP problem will occur in the process of classification. This paper adopts a middle way: It combines association rules with the ABN classification to construct a Bayesian classifier. It improves the independence assumption. Meanwhile, it avoids the NP problem. Finally we use our experimental results to show that it is better than the Naive Bayesian classifier in most fields.