Improving the Efficiency of Classifiers for Detecting Intrusion using Rough Set Theory

Abstract

This paper discusses the comparative analysis of different classifier algorithms for an Intrusion Detection System (IDS). There are numerous classification models available to classify the network activities as normal or abnormal. Over time, all the classifier algorithms are replaced by a better algorithm to overcome the new security threats and challenges faced by the network systems. For better accuracy and performance, network systems are moving to a better algorithm every day. Like all the anomaly detection algorithms, the most commonly applied Support Vector Machine (SVM), Neural Network (NN) and Fuzzy Logic (FL) models also have their own pits and falls. In addition, the use of a feature selection algorithm with these classifiers optimizes the performance and accuracy. The rough set based feature selection removes the irrelevant, redundant and noisy data. Even though these three models are known for their reduced rate of undetected anomaly, the higher false alarm rate and longer training time makes the user to think twice to apply these algorithms in a network system. These algorithms have pursued more knowledge and improving by the research conducted throughout the world every day. This paper analyses how these algorithms can be used in an Intrusion Detection System (IDS) with respects to their capabilities. The classifier algorithms have been implemented and evaluated by using standard benchmark KDD99 dataset. This paper also balances the advantages and disadvantages of these algorithms.