Real-time Machine Learning Based on Hoeffding Decision Trees for Jamming Detection in 5G New Radio

Real-time detection and mitigation of jamming attacks in 5G new radio (NR) is an urgent need. Over the past decade, jamming detection has been studied by many researchers using different methods including machine learning. However, nearly all the previous detection methods do not support real-time, which can result in delaying the mitigation and thereby degradation of the overall network performance. Therefore, in this paper, we propose a real-time jamming detection approach based on Hoeffding decision trees in 5G NR. First, we investigate the efficiency of decision trees in detecting jamming and use it as a baseline to validate the performance of the proposed approach. These models have been trained on a dataset built from 5G NR simulation, and the communication link is targeted by a malicious jammer. These models are then evaluated using metrics such as the probability of detection, probability of false alarm, accuracy, and training time. The preliminary results showed that HT can achieve an overall accuracy as high as 82% while traditional decision trees can achieve an accuracy as high as 100%. However, Hoeffding decision trees have the advantages of being trained online while traditional decision trees based models require the complete labeled dataset readily before training which delays the detection process.