Performance Evaluation of Artificial Immune System Algorithms for Intrusion Detection

Abstract

The need for real network traffic and large-scale realistic intrusion detection datasets in order to improve performances of IDS systems has make the use of Artificial immune system techniques to be on the rise. A dataset that contains benign and common attack network flows that mimics the real time can only help to train and test an intrusion detection system. In this paper, anomaly-based intrusion detection models are built using AIS algorithms namely AIRS1, Immunos, ClonalG and Dendritic Cell Algorithms. These algorithms were tested with NSL-KDD and CICIDS 2017 datasets, which have common updated set of malicious attacks such as DDoS, XSS, SQL Injection, Infiltration, Bruteforce, Portscan and Botnet. Our experiments show that AIS algorithms generally performs better both in accuracy and precision in detecting new attacks than other classifiers.