Research On Model Of Intrusion Detection System Based On Semi-supervised Classification

IDS(Intrusion Detection System) is an active defense system as the second line of network defense, has a non-substitutable function against other static defense system. As network attack mode being diversity, more and more intelligent technology is utilized in intrusion detection system. Semi - supervised learning is the hotspot of machine learning but its application in IDS is still very rare.This paper aims at the advantages and disadvantages of the IDS based on supervised learning and unsupervised learning, applies semi - supervised classification techniques to intrusion detection, according to the specific classification model, proposes a new algorithm and designs a semi - supervised intrusion detection model. First, it introduces the current status of intrusion detection systematically, sums up the problems and limitations existing in the current intrusion detection, and looking forward to the future trends. In view of the current semi - supervised learning being usually based on single classifier, since the integration of multiple classifiers has some incomparable advantages than the former. This paper researches mainly on semi - supervised classification based on ensemble learning, sums up theoretical basis of the ensemble learning, introduces some common ensemble learning algorithm and analyses preliminarily to the performance of ensemble learning. Finally, according to specific intrusion detection problem, put forward a semi - supervised algorithm suitable for intrusion detection dataset——RST(Regulization Self-Training), give an intrusion detection model based on semi - supervised classification——SSC(Semi-Supervised Classification) and verify the classification performance improvement of RST algorithm by some experiments.Semi - supervised classification of thesis design model and RST algorithm can fully reduce the quantity of the labels of network data artificially labelled by security experts, and makes full use of the unlabelled data, with some theoretical significance and value.