Intrusion Detection Method Based on Fuzzy Conditional Random Fields ?

Intrusion detection system is the indispensable part of every computer. With the increasing attack means, all kinds of intrusion detection methods have appeared. Compared with other intrusion detection methods, the intrusion detection methods based on Conditional Random Fields (CRFs) has better detection effect, but the problems that the accuracy is low when the training data is small and the training speed is very slow simultaneously exists. To solve these problems, this paper proposes a new intrusion detection method based on Fuzzy Conditional Random Fields (FCRFs). This method aims to integrate the fuzzy logic into CRFs, establish the objective function based on FCRFs, give the corresponding algorithm of parameter estimation and inference, and then explain how it is applied into intrusion detection. The experimental results on KDD cup 1999 intrusion data set show that this method can improve the training speed and can be superior in detection Precision, Recall, F-value and generalization capability to the well-known intrusion detection method based on CRFs. The improvement in attack detection Precision and training speed are quite obvious, particularly, for the U2R attack (36.7 percent and five times improvement in Precision and training speed).