Leveraging Lightweight Machine Learning for RF Jamming Detection in Mobile ad-hoc Networks: A Three-Tier Edge-Fog-Cloud Computing Approach

Abstract

Abstract Radio Frequency (RF) jamming is a prevalent threat in Mobile Ad-hoc NETworks (MANETs), yet effective detection on resource-constrained nodes remains challenging. This paper presents a distributed three-tier edge-fog-cloud framework leveraging lightweight machine learning for real-time detection across heterogeneous devices. A priority-based adaptation algorithm dynamically allocates computational tasks based on node capabilities. The framework achieves detection accuracies of 85.6% (edge/ lightweight Random Forest), 88.8% (fog/compressed Random Forest), and 97.8% (cloud/LightGBM). The dynamic adaptation mechanism enables efficient load distribution, reducing energy consumption by 66.9% compared to static cloud deployment with only an 8.2% accuracy cost. Cross-dataset validation on two independent physical-layer datasets – covering weak jamming and deceptive attacks under vehicular mobility (10-60 km/h) – demonstrates exceptional robustness with <1% mean accuracy variance, validating the framework’s effectiveness in dynamic, resource-constrained MANET environments.