Network Intrusion Detection Classification Based On Statistic Machine Learning

This paper studied the latest machine learning theories-manifold learning and Bayesian Yin-Yang learning on intrusion detection. The paper is focused on the following research topics.1. The classifier scheme combined with Isomap and one class support vector machine is proposed for U2R and R2L detection.2. Due to continuous and nominal attributes existed in TCP/IP connections, the algorithms of HL-Isomap and S-H-Isomap are proposed by replacing Euclidean distance with HVDM distance.3. A new method called k-variable method to construct connected neighborhood graphs both for LLE and Isomap algorithms is proposed. Robust kv-LLE and kv-Isomap are proposed. And the supervised S-kv-Isomap is proposed based on k-variable method.4. Linear manifold learning algorithms LPP and spectral regression (SR) algorithm to dimension reduction for intrusion detection are studied. The supervised classifier scheme LPP+RBF+k-NN and SR+GRNN+k-NN are proposed. SVM+k-NN fusion classifier is proposed and applied to the low dimension space by LPP.5. RPCL is one of method under the Bayesian Yin Yang learning frame. The clustering method RPCL+k-means and RPCL one class classifier are proposed.