Accurate path loss prediction directly impacts the fidelity of simulations and the quality of LoRa network planning decisions. Traditional models, such as Log-Distance and Okumura–Hata, are widely adopted but often fail to capture the propagation complexity of real-world environments. This work investigates the use of machine learning algorithms trained on path loss measurements from mobile nodes in urban scenarios to improve path loss estimation. Among the evaluated models, Random Forest and XGBoost achieved the lowest prediction errors, outperforming classical approaches. These models were integrated into the LoRaEnergySim framework and assessed under varying cell sizes, node densities, and spreading factors. Simulation results show that classical models generally overestimate losses, leading to higher energy consumption and unnecessary retransmissions, whereas tree-based models provide more accurate signal predictions and enable more efficient network configurations. Despite limitations related to dataset diversity, simulator assumptions, and computational resources, the findings highlight the benefits of incorporating machine learning-based propagation models into LoRa network design tools. Such integration improves simulation fidelity and supports more robust network planning. • ML reduces LoRa path loss prediction error by 30–40 • RF and XGBoost achieve 6.82–6.85 dB RMSE (112k urban samples). • ML propagation in LoRaEnergySim cuts simulated energy consumption by 42 • Classical models overestimate loss, causing inefficient LoRa planning. • Tree models scale better, improving packet delivery rates in large networks.
Ballestrin et al. (Fri,) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: