Model of Improved a Kernel Fast Learning Network Based on Intrusion Detection System

Abstract

The detection of network attacks on computer systems remains an attractive but challenging research scope. As network attackers keep changing their methods of attack execution to evade the deployed intrusion-detection systems (IDS), machine learning (ML) algorithms have been introduced to boost the performance of the IDS. The incorporation of a single parallel hidden layer feed-forward neural network to the Fast Learning Network (FLN) architecture gave rise to the improved Extreme Learning Machine (ELM). The input weights and hidden layer biases are randomly generated. In this paper, the particle swan optimization algorithm (PSO) was used to obtain an optimal set of initial parameters for Reduce Kernel FLN (RK-FLN), thus, creating an optimal RKFLN classifier named PSO-RKELM. The derived model was rigorously compared to four models, including basic ELM, basic FLN, Reduce Kernel ELM (RK-ELM), and RK-FLN. The approach was tested on the KDD Cup99 intrusion detection dataset and the results proved the proposed PSO-RKFLN as an accurate, reliable, and effective classification algorithm.