摘要： 半监督的双协同训练要求划分出的2个数据向量相互独立，不符合真实的网络入侵检测数据特征。为此，提出一种基于三协同训练(Tri-training)的入侵检测算法。使用大量未标记数据，通过3个分类器对检测结果进行循环迭代训练，避免交叉验证。仿真实验表明，在少量样本情况下，该算法的检测准确度比SVM Co-training算法提高了2.1%，并且随着循环次数的增加，其性能优势更加明显。

关键词: 入侵检测, 小样本, 支持向量机, 半监督, 双协同训练, 三协同训练

Abstract: The Co-training method requires the independence of two data vectors, which is far from the characteristic of real dataset in network intrusion detection. This paper proposes a intrusion detection method based on Tri-training. It exploits the large amount of unlabeled data, and increases the detection accuracy and stability by Co-training three classifiers. Simulation results show that this method is 2.1% more accurate than the SVM Co-training method, and it performs better with the increase of the loop number.

Key words: intrusion detection, small-sample, Support Vector Machine(SVM), semi-supervised, Co-training, Tri-training

中图分类号: 

• TP393.08