Research Of Intrusion Detection Based On Co-Training

With the popularity of computer networks, people take more and more attention to the network security. As a complement to traditional security mechanisms, intrusion detection gets more and more research. As network attack mode being diversity, more and more intelligent technology is utilized in intrusion detection system. Co-training is an important paradigm of semi-supervised learning, but its application in intrusion detection system is still very rare.This paper carries on the analysis and research to present intrusion detection system and variety of methods based on semi-supervised learning, applies Co-training algorithm to intrusion detection, and proposes an intrusion detection model based on Co-training. Firstly, 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. And then it totally introduces the method of ensemble learning and semi - supervised classification. Finally, it studies the principles, process and two important algorithms Tri-training and Co-Forest of Co-training.Finally the experiment data carry on the sampling from KDD Cup 99’s data set, and takes intrusion detection rate and the rate of false alarmas as the standard to examine algorithm performance. The results show that intrusion detection based on Co-training can achieve a higher detection rate and low false alarm rate in terms of less labeled data. Meanwhile, this method can take advantage of a large number of unlabeled data effectively, reduce the dependence on labeled data, and it has certain theory significance and the practical value.