Optimisation de Routage dans les Réseaux VANETparApprentissage Supervisé

Abstract

This thesis focuses on optimizing routing in Vehicular Ad Hoc Networks (VA NET), a key component of Intelligent Transportation Systems (ITS). These dynamic networks enable vehicles to exchange real-time data (V2V and V2I), thereby enhan cing road safety and improving traffic management. However, their dynamic nature, unstable topology, and high node mobility complicate data routing, adversely affecting the quality of service (QoS). To address these challenges, we propose the integration of Artificial Intelligence (AI), specifically supervised machine learning, into the HRLB SVDN protocol. The chosen model, CatBoostClassifier, can effectively analyze hetero geneous data with high precision. Trained on realistic simulated datasets, it predicts optimal communication paths based on parameters such as vehicle density, inter-node distance, and link stability. The experimental results are highly promising : an accu racy of 98.75%, a ROC-AUC score of 99.79%, an average precision of 99.71%, and a logarithmic loss of just 0.038. These metrics demonstrate a strong ability to identify reliable routing paths in dynamic network environments. Simulations also revealed a significant improvement in protocol performance, with a 10.75% increase in average delivery rate, a 7.86% increase in throughput, and a 10.08% reduction in average la tency. In conclusion, this work highlights the effectiveness of AI-based approaches in designing intelligent, robust, and adaptive routing protocols tailored to the demanding requirements of modern VANET environments.