Abstract

Vehicular Ad Hoc Networks (VANETs) have emerged as a critical component in intelligent transportation systems (ITS), enabling vehicle-to-vehicle (V2V) and vehicle-to-infrastructure (V2I) communication. However, the highly dynamic topology, high mobility, and low latency requirements of VANETs present significant challenges for ensuring reliable and efficient data transmission. Traditional machine learning models often struggle to adapt to VANETs’ real-time data processing needs and variable network conditions. While deep learning offers promising capabilities in feature extraction and pattern recognition, standalone architectures may fall short due to overfitting, underfitting, or limited generalization in complex VANET environments. This study proposes an improvised ensemble deep learning framework that integrates Convolutional Neural Networks (CNN), Recurrent Neural Networks (RNN), and Transformer-based attention mechanisms. The ensemble model leverages the spatial-temporal feature extraction strength of CNN-RNN and the long-range dependency modeling capability of Transformers. A weighted majority voting and adaptive fusion layer are implemented to combine model outputs effectively. The framework is evaluated using real-time vehicular mobility datasets and simulated traffic scenarios to measure metrics such as packet delivery ratio (PDR), end-to-end delay, and throughput. The proposed ensemble framework achieved a 15–20% improvement in PDR, a 25% reduction in end-to-end delay, and a significant increase in throughput compared to existing deep learning baselines.

Authors

Sesham Anand¹, Pitty Nagarjuna²
Maturi Venkata Subba Rao Engineering College, India¹, Indian Institute of Science, Bengaluru, India ²

Keywords

VANETs, Ensemble Deep Learning, Vehicular Communication, CNN-RNN, Transformer Attention