Machine Learning-Based Attack Detection for the Internet of Things

Abstract

The number of Internet of Things (IoT) device connections is increasing rapidly as IoT applications are vital in any operation. IoT must maintain safe internet access that withstands various malicious attacks for instance Recon, Mirai, Distributed Denial of Service (DDoS), and Spoofing which has gained much attention. Intelligently changing and zero-day attacks are emerging every day. This highlights the need for intelligent security solutions tailored specifically to this technology. Various Machine Learning (ML) based approaches have been utilized for intrusion detection to tackle IoT attacks. However, the flaws of current attack detection and feature extraction techniques result in low detection accuracy. Thus, it hindered their real-world applications and highlighted the need fora lightweight and computationally robust model trained and assessed on a recent datasets. Therefore, this work proposed an attack detection model trained and validated using the CICIoT2023 and CICIDS2017 datasets. Initially, data preprocessing is done then features are extracted by using an unsupervised Elastic Deep Autoencoder (EDA) with optimum hyperparameters. Further, the Extreme Gradient Boosting (XGBoost) binary classifier is tuned by the Grey Wolf Optimizer (GWO) and fed extracted feature sets to classify attacks. The results of the experiments show the effectiveness of our model with a higher detection accuracy in both datasets. Finally, the performance comparison confirmed that the results of the proposed work is competitive with other state-of-the-art method in securing IoT infrastructures.