Abstract—In recent years, we have witnessed rapid growth in the application of IoT globally. IoT has found its applications in governmental and non-governmental institutions. The integration of a large number of electronic devices exposes IoT technologies to various forms of cyber-attacks. Cybercriminals have shifted their focus to the IoT as it provides a broad network intrusion surface area. To better protect IoT devices, we need intelligent intrusion detection systems. This work proposes a hybrid detection system based on Genetic Algorithm (GA) and Extreme Learning Method (ELM). The main limitation of ELM is that the initial parameters (weights and biases) are chosen randomly affecting the algorithm’s performance. To overcome this challenge, GA is used for the selection of the input weights. In addition, the choice of activation function is key for the optimal performance of a model. In this work, we have used different activation functions to demonstrate the importance of activation functions in the construction of GA-ELM. The proposed model was evaluated using the TON_IoT network data set. This data set is an up-to-date heterogeneous data set that captures the sophisticated cyber threats in the IoT environment. The results show that the GA-ELM model has a high accuracy compared to single ELM. In addition, Relu outperformed other activation functions, and this can be attributed to the fact that it is known to have fast learning capabilities and solves the challenge of vanishing gradient witnessed in the sigmoid activation function.
 
Keywords—intrusion detection system, extreme learning machine, genetic algorithm, TON_IoT data sets, hybrid


Copyright © 2023 by the authors. This is an open access article distributed under the Creative Commons Attribution License (CC BY-NC-ND 4.0), which permits use, distribution and reproduction in any medium, provided that the article is properly cited, the use is non-commercial and no modifications or adaptations are made.