A comparative analysis of machine learning models based on weighted input parameters for V2V path loss prediction in highway, rural, suburban, and urban environments

Vehicle-to-vehicle (V2V) communication plays a crucial role in intelligent transportation systems by enhancing safety, efficiency, and connectivity in mobility. However, the accuracy of path loss prediction in communication channels is significantly affected by varying and complex propagation environments. This study conducts a detailed analysis of machine learning-based models to improve path loss prediction in V2V communication. A dataset containing 161,940 data points collected from rural, highway, suburban, and urban environments was used to evaluate different machine learning algorithms. For this purpose, AdaBoost, Random Forest, Artificial Neural Networks, Support Vector Regression, and Gradient Boosting models were trained using environmental and system parameters such as distance, obstacle types, modulation schemes, and weather conditions. Additionally, comparative analyses were conducted against traditional empirical models, including log-distance, two-ray, and log-ray methods. The results demonstrate that machine learning models significantly outperform traditional methods across all environments. Specifically, AdaBoost achieved the best performance in highway and rural environments, with RMSE values of 0.00776 and 0.00124, and R${}^2$ values of 0.99965 and 0.99935, respectively. Random Forest provided the best results in suburban and urban environments, with RMSE values of 0.01274 and 0.01369, and R${}^2$ values of 0.99952 and 0.99938, respectively. Moreover, assigning weighted importance to environmental parameters substantially improved model performance. Overall, this study highlights the efficiency and necessity of machine learning-based path loss prediction in V2V communication. The findings indicate that machine learning models dynamically learn the effects of environmental factors, providing more reliable and efficient predictions compared to traditional models. Furthermore, this research offers valuable insights for improving traffic flow and reducing accidents in traffic-congested areas.