A Deployment-Aware Feasibility Framework for Machine Learning-Based IoT Intrusion Detection Across Edge, Fog, and Cloud Architectures
En palabras de los autores
The rapid growth and heterogeneity of Internet of Things (IoT) environments have exposed fundamental limitations in traditional rule-based and signature-based intrusion detection systems. This paper presents a quantitative deployment-aware analysis of machine learning (ML)-based intrusion detection approaches across edge, fog/gateway, and cloud architectures. Unlike prior surveys that primarily emphasize detection accuracy, this work defines representative quantitative deployment capability envelopes extracted from experimental and system-level studies and introduces a structured Deployment Feasibility Score (DFS) model. The proposed framework maps ML techniques to architectural layers based on computational demand, memory footprint, and latency sensitivity using a weighted ordinal scoring mechanism. The analysis demonstrates that lightweight statistical and linear models are most suitable for edge deployment, ensemble and clustering-based methods align with fog/gateway environments, while deep and optimization-driven models are best suited for cloud infrastructures. By formalizing deployment feasibility through quantitative grounding and structured evaluation, this work provides practical guidance for selecting intrusion detection solutions under real-world architectural constraints, supporting more informed and deployment-conscious IoT security design.
Apareció: jueves, 8 de octubre. arXiv. Preprint con versión publicada.
DOI: 10.1109/CSNT69054.2026.11502111
Versión publicada: 2026 IEEE 15th International Conference on Communication Systems and Network Technologies (CSNT), pp. 959-966, 2026
Comentario de los autores: 8 pages, 3 figures, 7 tables. Published in the 2026 IEEE 15th International Conference on Communication Systems and Network Technologies (CSNT)