Exploring Machine Learning Techniques for Path Loss Prediction in LoRa Networks

This study investigates the performance of various machine learning techniques in predicting signal losses in Long Range (LoRa) networks, a radio solution for Internet of Things (IoT) applications. Utilizing an existing dataset containing propagation parameters, models such as decision tree, random forest, support vector regression (SVR), Lasso regression, XGBoost, and neural networks were trained and evaluated. Each model was chosen for its unique advantages in capturing the complex relationship between input variables and signal losses in LoRa mobile node propagation. Through comprehensive analysis, it was observed that, in terms of Root Mean Square Error (RMSE), all proposed AI models outperformed traditional algorithms, except for SVR with a sigmoid kernel. The topperforming AI models demonstrated a significant reduction in RMSE compared to the Log-Distance model, which was the algorithm with the smallest RMSE in the study from which the dataset was sourced. Although AI-based models may entail higher computational overhead, the enhanced prediction accuracy justifies the additional computational expense. These findings provide insights into optimizing the reliability and performance of LoRa networks, thereby facilitating the deployment of IoT applications across diverse real-world scenarios.