Abstract:

to high dimensional co-linearity problem of network intrusion anomaly detection model’s input features and dynamic changes of network environment, a new fast anomaly detection model construction approach based on dimension reduction strategy using principal component analysis (PCA) and mutual information (MI) is proposed in this paper. At first, PCA based feature extraction method is used to extract independence latent features, to diminish co-linearity among these input variables. Then, MI based feature selection method is used to select important features from PCA extracted latent features. Thus, these independent features that have much relation to anomaly detection model’s output are selected. At last, a kind of machine learning algorithm with fast learning speed, i.e., random vector function link (RVFL) net, is used to construct the final intrusion detection model with these extract and selected features. Simulation results based on KDD99 data set show that the proposed method can extract and select features effectively with fast learning speed.

Key words: network intrusion, anomaly detection, dimension reduction, machine learning