Feature selection in intrusion detection, state of the art: A review

With the increase of internet usage the need of security for organizations network also increased. Network anomaly intrusion detection systems are designed to monitor abnormal activity in the network. These systems find the behavior that is deviated from the normal behavior. Network anomaly detection methods are implemented using different approaches including machine learning, data mining, and many more. However, intrusion detection systems highly depend on the features of the input data. These input features give information to the learning algorithms which used in intrusion detection system in the form of the detection method. With irrelevant and redundant features learning algorithm builds detection model with less accuracy rate. Also, ambiguous features increase the time complexity and consume other computational resources as well. By removing these irrelevant and redundant features accuracy of the learning algorithms can be increased. In this paper implementation of different feature selection techniques have been reviewed. Novel feature selection techniques have been developed due to its importance in network intrusion domain. We have discussed some of it in a technical aspect. These techniques are being discussed in detail. Moreover, features from these methods are also given and their results are being. We categorized these techniques according to their implementation. Different comparison of these techniques have been given and been discussed. Moreover, the benchmark dataset that is KDD99 widely used for anomaly detection is also discussed in this paper. © 2005 - 2016 JATIT & LLS. All rights reserved.