Study of meta heuristic algorithms for network intrusion detection

Abstract

In recent days, Information exchange has become an inevitable necessity. Almost all the fields require information for processing. Information is being stored in the Information Systems (ISs). So, attacks over ISs are increasing day by day. In order to protect the ISs, from malicious intruding activities, intrusion detection systems (IDSs) are to be deployed. IDSs will monitor the traffic flow inside the network and classify the packets as malicious or non-malicious and alert the security professional accordingly. The crux of this research work is to build an effective IDS framework which is capable of producing high accuracy and low False Positive Rate (FPR) while analyzing the packets. Combination of feature selection techniques such as filter (Chi-square algorithm) and wrapper based algorithms are utilized along with Meta Heuristic algorithms such as Whale Optimization, Bat and Cuckoo Search algorithms. The best features selected through these algorithms hybridization are passed into a decision tree classifier algorithm to find the prediction measures of the attacks found in the NSL-KDD Dataset. While considering the results produced by the chi-square and bat algorithm in combination with the decision tree algorithm (DT_CHI_Bat), 14 predominant features were chosen with 22,543 traffic record instances and 99.04 % accuracy was achieved. Also, the time taken to make this accuracy prediction is 0.63 seconds. The achieved results through DT_CHI_Bat are superior in this perspective as compared to DT_CHI_Whale, DT_CHI_Cuckoo and other state of the art approaches. newline