Towards DoS Attack Detection for IoT Systems: A Cross-Layer Oriented Approach Based on Machine Learning Techniques
Dimitrios Tasiopoulos, Apostolos Xenakis, Alexios Lekidis, Dimitrios Kosmanos, Costas Chaikalis, Vasileios Vlachos · 2025
The rapid expansion of Internet of Things (IoT) systems has expanded the threat surface as well as the increased likelihood of cyber-attacks that can result in Denial-of-Service (DoS) in the system operation. Prominent attacks leading to DoS belong to the blackhole, flooding, sinkhole, and spoofing categories. The detection of these attacks though still remains a great challenge, which necessitates the presence of more accurate and robust detection methods. To this end, in this paper we investigate the use of Machine Learning (ML) models to perform cross-layer detection and compare them with singlelayer detection approaches across several IoT systems. In our analysis, we evaluate performance metrics (accuracy, precision, recall, F1-score) across all attacks. Results reveal that crosslayer methodologies consistently outperform single-layer detection, achieving higher accuracy and almost all cases lower variability. Additionally, feature selection is another advantage of cross-layer methodologies, with multi-layered datasets enabling more robust detection compared to single-layer approaches. Hence, cross-layer frameworks overall allow enhancing security in multi-threat IoT environments.