基于RVM的网络流量分类研究

将相关向量机(RVM)分类模型应用于网络流量分类问题中. 首先对实验数据进行了标准化处理, 然后将RVM与其他机器学习算法进行了性能比较, 最后在RVM分类结果预测概率中引入置疑区间概念, 研究了置疑区间范围及其对分类准确性的影响, 并基于此提出了一种新的混合流量分类方法. 实验结果表明: 1) RVM在准确性等3方面性能指标上优于SVM, 且在小样本情况下仍具有较高的分类准确率; 2) 置疑区间[0.1,0.9]内的分类预测准确率较低, 而置疑区间之外的分类预测准确率在98%以上.

Network Traffic Classification Based on Relevant Vector Machine

Relevant vector machine (RVM) is applied in network traffic classification. Firstly, experiment data is standardized, and then RVM is compared with other machine learning tools. Lastly, doubting interval is introduced to analyze predicted probability of classification, based on which a new hybrid traffic classification approach is proposed. Experiment studies illustrate that: 1) RVM excels the support vector machine (SVM) in three performances, and moreover, its classification accuracy is rather high in the situation of small sample circumstances; 2) probabilistic classification in doubting interval has a rather low classification accuracy while an accuracy above 98% outside doubting interval.