An Intelligent Collaborative Edge–Cloud Deep Learning Framework for Secure and Real-Time Anomaly Detection in Internet of Things Networks

Abstract

Internet of Things (IoT) networks connections reached 21 Billion across the globe [4]. As their population grows, the attacks also increases. They are susceptible for various cyber-attacks such as Botnet, DDoS, MitM, Firmwere, Encryption, Brute force etc., To prevent IoT networks from such attacks, the Intrusion Detection Systems (IDS) are used. Three types of IDS prominently used in the industry. Cloud based IDS, Edge Computing based IDS and Deep Learning based IDS. Cloud approach experiences latency, Edge Computing approach experiences resource constraints while the Deep Learning approach need high computation capability. This article proposes a novel Collaborative Edge–Cloud Deep Learning Framework (EC-DLF) to reap the benefits of three design approaches. The pruning and knowledge distillation helps IoT devices to handle the intrusion detection with the less resource of about < 2MB RAM with the very less latency of <4ms. The cloud approach used in this work uses 6 layer encoder with multi heads, which tested on 4 different datasets namely, N-BaIoT, TON_IoT, Edge-IIoTset, and CICIoT2023. The results were impressive. F1 Score of 99.43%, AUC ROC of 0.9979 and MCC of 0.9932. These results were compared with five baseline models and the proposed model outperformed all the baseline models. Further the statistical analysis also done to confirm the robustness of the system. Wilcoxon signed-rank test was used, which yields the p < 0.001 and the Cohen’s d > 4.2. On the system performance aspect, t-SNE analysis, throughput, cross data validation, scalability analysis also performed.