Intrusion-Miner: A Hybrid Classifier for Intrusion Detection using Data Mining

Abstract

With the rapid growth and usage of internet, number of network attacks have increase dramatically within the past few years. The problem facing in nowadays is to observe these attacks efficiently for security concerns because of the value of data. Consequently, it is important to monitor and handle these attacks and intrusion detection system (IDS) has potentially diagnostic ability to handle these attacks to secure the network. Numerous intrusion detection approaches are presented but the main hindrance is their performance which can be improved by increasing detection rate as well as decreasing false positive rates. Optimizing the performance of IDS is very serious issue and challenging fact that gets more attention from the research community. In this paper, we proposed a hybrid classification approach ‘Intrusion-Miner’ with the help of two classifier algorithm for network anomaly detection to get optimum result and make it possible to detect network attacks. Thus, principal component analysis (PCA) and Fisher Discriminant Ratio (FDR) have been implemented for the feature selection and noise removal. This hybrid approach is compared with J48, Bayesnet, JRip, SMO, IBK and evaluate the performance using KDD99 dataset. Experimental result revealed that the precision of the proposed approach is measured as 96.1 % with low false positive and high false negative rate as compare to other state-of-the-art algorithm. The simulation result evaluation shows that perceptible progress and real-time intrusion detection can be attained as we apply the suggested models to identify diverse kinds of network attacks.