Enhancing LoRaWAN Performance Using Boosting Machine Learning Algorithms Under Environmental Variations.

Area of Science:

• Wireless Communication
• Machine Learning
• Environmental Sensing

Background:

• Path loss prediction is critical for Long-Range Wide-Area Network (LoRaWAN) optimization.
• Existing Machine Learning (ML) models often overlook dynamic environmental factors like temperature, humidity, pressure, and particulate matter.

Purpose of the Study:

• To evaluate the performance of five boosting ML models (AdaBoost, XGBoost, LightGBM, GentleBoost, LogitBoost) for LoRaWAN path loss prediction under varying environmental conditions.
• To compare these models against theoretical approaches and previous studies using metrics like RMSE, MAE, and R².
• To analyze the trade-off between model accuracy and computational complexity (training time, inference latency, model size, energy consumption).

Main Methods:

• Implementation and evaluation of five boosting ML algorithms: AdaBoost, XGBoost, LightGBM, GentleBoost, and LogitBoost.
• Comparison with Log-Distance and Okumura-Hata theoretical models.
• Hyperparameter tuning using Bayesian Optimization.
• Performance assessment using Root Mean Square Error (RMSE), Mean Absolute Error (MAE), and R².
• Computational complexity analysis including training time, inference latency, model size, and energy consumption.

Main Results:

• Barometric pressure was identified as the most significant environmental factor influencing path loss across all evaluated models.
• LightGBM demonstrated superior performance, achieving the lowest RMSE (0.5166) and highest R² (0.7151).
• LightGBM provided the optimal balance between predictive accuracy and computational efficiency.

Conclusions:

• Boosting algorithms, particularly LightGBM, are highly effective for accurate path loss prediction in LoRaWAN environments, even under dynamic environmental conditions.
• Incorporating environmental factors like barometric pressure significantly enhances prediction accuracy.
• LightGBM offers a compelling solution for efficient and accurate LoRaWAN path loss modeling.