A Lightweight Misbehavior Detection System for Communication Based Train Control (CBTC) Systems Based on Ensemble Learning

As a core technology in modern metro rail systems, Communication Based Train Control (CBTC) plays a critical role in ensuring safe, efficient, and real-time train operations. However, intrusion detection systems in CBTC environments may encounter challenges related to robustness and response time, particularly when processing highly imbalanced data. This study proposes a lightweight misbehavior detection system (MDS) for CBTC using ensemble learning. Bagging methods such as Random Forest and k-nearest Neighbors, along with boosting methods like AdaBoost, XGBoost, and LightGBM, were tested on the CBTCSet dataset by addressing data imbalance and real time requirements. Random Forest with a weighted strategy achieved the best balance, with 92 percent accuracy, precision, recall, F1 score, and an average processing time of 70.23 microseconds per entry. These results support the use of ensemble methods, especially Random Forest, as an effective and efficient solution for real-time MDS in CBTC systems.