Machine Learning-Based Cybersecurity in Advanced Autonomous and Connected Vehicles

Abstract—The rapid growth of Advanced Autonomous and Connected Vehicles (AACVs) has been revolutionizing transport while building significant cybersecurity risks. AACVs relies on Electronic Control Units (ECUs), networked sensors, vehicle-to-everything (V2X) networks, and cloud facilities, which makes them vulnerable to replay attacks, GPS spoofing, Man-in-the-Middle (MITM) attacks, malware injection, and denial-of-service (DoS) attacks. Smart key systems have been compromised using software-defined radio (SDR) tools such as HackRF One to offer unauthorized access through RF signal replay. Similarly, laser interference with LiDAR and camera sensors has shown the capability to disrupt autonomous navigation. To address such threats, machine learning (ML)-based techniques are employed for anomaly detection, predictive threat analysis, and intrusion prevention. The proposed framework uses decision trees, ensemble models, and generative adversarial networks (GANs) in combination to detect cyberattacks in real time. Federated learning is utilized to preserve data privacy through facilitating joint model training on various vehicles without sharing sensitive raw data with central servers.An intelligent multi-stage intrusion detection system (IDS) is utilized, combining rule-based filtering and ML classifiers for low-latency, high-accuracy threat detection. Experimental evidence indicates that the use of ML for anomaly detection reduces the impacts of replay attacks, network intrusion, and spoofing of sensors. Other safeguards like secure diagnostic protocols, biometric authentication, and encryption are incorporated to deter zero-day attacks in AACVs.