Urban Path Loss Modelling: Integrating Machine Learning and Geometry Extraction

In this study, we present a comprehensive system for path loss prediction that integrates data from both real measurement campaigns and ray tracing simulations for outdoor scenarios. By leveraging the strengths of LightGBM, a powerful gradient boosting framework, optimized through Optuna, an advanced hyperparameter optimization tool, we aim to enhance the model's accuracy and efficiency. The system pipeline encompasses several stages, including setting up scenarios, feature extraction, preprocessing, and model training, followed by thorough performance evaluation. Our findings indicate that the optimized LightGBM model significantly outperforms traditional empirical models and with much less resources as ray tracing, achieving lower MSE and RMSE values and higher R2 scores. The high R2 score, in particular, demonstrates the model's ability to explain a large proportion of the variance in path loss, underscoring its predictive power and robustness. Additionally, the integration of diverse data sources and advanced optimization techniques ensures a more accurate and stable prediction of path loss through different unseen environments, which is crucial for the design and optimization of wireless communication systems.