基于关联规则的贝叶斯网络分类器

Bayesian network classifier with association rules

关联规则分类器(CBA)利用关联规则来构造分类算法,但其没有考虑分类问题中的不确定性。提出一种基于关联规则的贝叶斯网络分类算法。该算法利用关联规则挖掘算法提取初始的候选网络边集,通过贪心算法学习网络结构,得到比经典的贝叶斯网络分类器TAN更好的拓扑结构。通过在15个UCI数据集上的实验结果表明,该算法取得了比TAN,CBA更好的分类性能。

Classification Based on Association （CBA） algorithm built a classifier based on the association rules, without considering the uncertainty in the classification problem. This paper proposed a Bayesian network classifier based on the association rules. The improved algorithm initialized the graph structure with candidate edges, which was extracted by Apriori algorithm, and obtained a better network structure than Tree-Augmented Network （TAN） classifier with the greedy search. The empirical study on 15 UCI datasets shows that the improved algorithm yields higher accuracies than TAN, CBA.