Real-Time Forest Fire Risk Assessment in Tunisia: Integrating LoRaWAN IoT with Machine Learning and FWI

Forest fire risk prediction is a critical challenge in environmental monitoring, particularly in Mediterranean ecosystems increasingly affected by extreme climatic conditions. This paper presents an operational forest fire prediction framework based on machine learning for estimating the Fire Weather Index (FWI), a widely used indicator of fire danger derived from meteorological variables. Several regression models were trained and evaluated using a historical fire-weather dataset from the UCI Machine Learning Repository to identify the most accurate predictive approach. Following comprehensive preprocessing, feature selection, and hyperparameter optimization, tree-based ensemble models especially XGBoost demonstrated strong predictive performance and robust generalization capability. The trained model was subsequently integrated into a real time prediction system that combines a LoRaWAN-enabled IoT sensing infrastructure with a web-based deployment platform. Real-time environmental measurements collected by IoT sensors are processed using the same preprocessing pipeline and fed into the trained model to enable continuous online FWI estimation. The proposed framework demonstrates the practical feasibility of deploying optimized machine learning models for real-time forest fire risk monitoring and early warning applications.