Enhancing IoT Security Through Intelligent Key Compromise Detection: A Conv1D-Based Framework for SDN-Fog Networks | ||||
| Engineering Research Journal (Shoubra) | ||||
| Articles in Press, Accepted Manuscript, Available Online from 22 July 2025 | ||||
| Document Type: Research articles | ||||
| DOI: 10.21608/erjsh.2025.399506.1431 | ||||
 View on SCiNiTO | ||||
| Authors | ||||
Eman Omar Soliman  1; Hala Mansour2; Heba Allah Adly TagElDien2; Shimaa Salama2
| ||||
| 1National Telecommunications Institute, Cairo, Egypt | ||||
| 2Electrical Engineering Department, Faculty of Engineering at Shoubra, Benha university, Cairo, Egypt. | ||||
| Abstract | ||||
| The pervasive expansion of the Internet of Things (IoT) necessitates the development of sophisticated security paradigms capable of countering advanced cyber threats, particularly those targeting the compromise of cryptographic keys within Fog of Things (FoT) infrastructures. This paper presents an in-depth comparative analysis of four prominent machine learning and deep learning models—specifically, a one-dimensional Convolutional Neural Network (Conv1D), an Autoencoder-based anomaly detector (AE), Random Forest (RF), and Extreme Gradient Boosting (XGBoost)—evaluated for their effectiveness in identifying compromised key attacks using the comprehensive CIC-ToN-IoT dataset. We assessed the performance of these models in both binary anomaly detection (distinguishing normal traffic from attacks) and multi-class classification scenarios (identifying specific attack types such as backdoor, injection, password, and ransomware). Our experimental findings reveal the superior capability of the Conv1D model, which achieved an outstanding accuracy of 99.16% in binary detection and 99.98% in multi-class classification, coupled with remarkably low false positive and false negative rates. The robustness and generalizability of the models were rigorously validated through k-fold cross-validation, label permutation tests, and assessments of resilience against noise injection, confirming their stability under varied conditions. Furthermore, analysis of inference latency highlights the practical feasibility of deploying these models in real-time within Software-Defined Networking (SDN)-enabled fog computing environments to secure IoT ecosystems against the critical threat of cryptographic key compromises, offering significant contributions to the field of network security in emerging FoT architectures. | ||||
| Keywords | ||||
| Fog of Things; Deep Learning; Attack Classification | ||||
|
Statistics Article View: 32 |
||||
