Ensemble Technique Based on Supervised and Unsupervised Learning Approach for Intrusion Detection

Abstract

Security of networks within an organization is one of the most crucial issue for any organizations. Numerous techniques have either been developed or implemented to secure computer network and communication over the Internet. One method that has gathered attention under security domain over the years is the Intrusion detection method. This security technique analyzes information from various nodes within a network to identify possible threat. In this paper an ensemble technique using supervised and unsupervised learning approach has been proposed. At first Clustering is performed over data and then classification of data is performed. Clustering is used so as to detect unknown attacks in the networks and also to form clusters of same type of data. Then with the help of Classification algorithms classification of data into its appropriate classes is done and it is also used to measure the detection rate, false positive rate etc. NSL-KDD, KDD Cup’99 and Kyoto 2006+ datasets are used in this paper for experimentation purpose. The results of misuse-based intrusion detection and proposed system is compared on various parameters like detection rates, false positive rates, precision, true positive rate. Results prove that the proposed approach has better low false positive as well as detection rates, than misuse based intrusion detection. Proposed System detects various types of attacks with high percentage of detection as well as and low false positive rates. This system is also compared with existing systems which were described in research papers and results shows that our system gives less false positive rates than existing systems.