Smart intrusion detection model to identify unknown attacks for improved road safety and management

The Internet of Vehicles (IoV) has emerged as a transformative technology for intelligent transportation systems, enabling real-time communication between vehicles, infrastructure and external networks. However, this connectivity also introduces significant cybersecurity risks, such as spoofing, injection and denial of service (DoS) attacks, which threaten operational safety and system reliability. To address these challenges, this study proposes the adaptive CNN-based intrusion detection system (ACIDS), a robust and scalable framework designed to enhance intrusion detection in IoV environments. ACIDS integrates convolutional neural networks (CNN) for hierarchical feature extraction, the synthetic minority over-sampling technique (SMOTE) to address class imbalance and an open-set classification framework to detect novel attack patterns. The model was evaluated on the AWID dataset, achieving an accuracy of 94%, a perfect detection rate of 100% and a low false alarm rate of 3%. Additional validation on the NSL-KDD dataset confirmed its generalizability, with an accuracy of 91.7% and a detection rate of 98.3%. These results significantly outperform baseline models, including support vector machines (SVM) and random forests (RF), as well as recent methods such as transformer-based and hybrid RNN-CNN approaches. Key parameters used for benchmarking include accuracy, detection rate, false alarm rate, precision, F1-Score and AUC-ROC, demonstrating the model's balanced performance and computational efficiency. By addressing critical issues such as class imbalance, adaptability to novel threats and real-time scalability, ACIDS offers a practical solution for securing IoV systems. Its low computational overhead and ability to operate on resource-constrained edge devices further emphasize its suitability for real-world deployments.