基于贝叶斯学习的集成流量分类方法

摘要

It is difficult to model with the conditional independence assumptions of Naive Bayes(NB) method and the small training set of Bayes Network Augmented Naive Bayes(BAN) approach. In order to solve this problem, a new classification method is proposed in this paper. This is a combined traffic classification based on instance-based learning. It constructs a separate NB and BAN classifiers and obtains each classifier weight according to the validation set. It obtains the classification of network traffic through weighted average combination of classifier output. Using Moore data set as the experimental data, results show that the ensemble learning method rather than NB method and BAN method has higher classification accuracy and stability.%NB方法条件独立性假设和BAN方法小训练集难以建模.为此,提出一种基于贝叶斯学习的集成流量分类方法.构造单独的NB和BAN分类器,在此基础上利用验证集得到各分类器的权重,通过加权平均组合各分类器的输出,实现网络流量分类.以Moore数据集为实验数据,并与NB方法和BAN方法相比较,结果表明,该方法具有更高的分类准确率和稳定性.