Novel RF-jamming IDS for vehicular networks: LSTM, XGBoost, and meta-models approaches

The rapid advancement of Intelligent Transportation Systems (ITS) has facilitated diverse safety and comfort services via vehicular networks. However, despite these technological improvements, vehicular networks remain exposed to Radio Frequency (RF) jamming attacks, which continue to challenge the effectiveness of Intrusion Detection Systems (IDS). This paper addresses this topic by proposing novel IDS architectures based on Long Short-Term Memory (LSTM) and eXtreme Gradient Boosting (XGBoost), as well as meta-models that integrate these methods to enhance detection and classification. Through a comprehensive analysis of several models, this research demonstrates that the best LSTM model, LSTM_RELU_IO-L1, achieves an accuracy of 95.67% and 85.17% for vehicles traveling at 25m/s and 15m/s, respectively. These results confirm its superior performance compared to baseline and recent models: K-Nearest Neighbors (94.46%/82.27%), Random Forest (94.61%/80.04%), and a previous LSTM work (95.17%/84.83%). Furthermore, the proposed RFV-LG meta-model, which combines LSTM and XGBoost, shows substantial improvements. Particularly, the RFV-LG-H6 model reaches 96.46% accuracy for low-speed vehicles, which is the most critical scenario for IDS, and nearly 100% accuracy for high-speed vehicles. These proposals advance the state-of-the-art RF jamming IDS for vehicular networks while establishing the RF jamming IDS meta-models as cutting-edge solutions with strong potential for real-world deployment.