灵活的增强朴素贝叶斯分类器

Flexible Augmented Naive Bayesian Classifier

摘要: 提出了一种新颖的基于最小描述长度（Minimum Description Length，MDL）准则的灵活的增强朴素贝叶斯分类器（Flexible Augmented Naive Bayesian classifier，FAN）算法．该算法能够根据数据集自适应地匹配从朴素贝叶斯分类器（Naive Bayesian classifier，NB）到树增强朴素贝叶斯分类器（Tree Augmented NaiveBayesian classifier，TAN）的网络结构，且保持了TAN计算简单和鲁棒性的特点．在UCI数据集上用分层交叉验证的方法对NB、TAN、FAN算法进行测试，实验结果表明FAN算法具有良好的分类精度．

Abstract: A new flexible augmented naive Bayesian classifier(FAN) algorithm based on the minimum description length(MDL) rule is proposed.According to data sets,this algorithm is able to adapt itself to match network structures from naive Bayesian classifiers(NB) to tree augmented naive classifiers(TAN),and maintains the computational simplicity and robustness that characterize TAN.NB,TAN and FAN are tested by stratification-cross-validation on the sets of UCI,and the experiment results show that the FAN algorithm holds a good classification accuracy.