Design And Development of an Intelligent Fire Risk Prediction System Based on Internet of Things (IOT) Technologies
No Thumbnail Available
Date
2026
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
The escalating global wildfire crisis, intensified by climate change, presents a critical
socio-ecological threat in Mediterranean regions such as Algeria. Traditional
monitoring systems, reliant on reactive firefighting and static risk maps, coupled
with opaque artificial intelligence models, are demonstrably inadequate for proactive,
trustworthy risk management. This doctoral research confronts these limitations
by developing a novel, trans-disciplinary framework for real-time, interpretable forest
fire risk prediction and mitigation.
The core innovation is the synergistic integration of a robust, edge-deployed IoT
sensor network with a multi-faceted hybrid intelligence engine. This architecture
bridges critical operational gaps: a LoRaWAN-based Wireless Sensor Network acquires
hyper-local data; a tripartite Explainable AI (XAI) framework (SHAP, BorutaSHAP,
LIME) ensures model transparency; and a hybrid modeling core strategically
partitions computation across an edge-cloud continuum for efficiency.
Methodologically, the framework integrates ensemble classifiers (e.g., XGBoost,
CatBoost), deep learning forecasters (LSTM, hybrid RNN-LSTM, NeuralProphet),
Bayesian models, and a hierarchical fuzzy-logic system. A pivotal advancement is the
incorporation of physics-informed regularization, which grounds predictions in fire
behavior physics. The system was validated through comparative analysis across regions
and longitudinal prototype testing in Khenchela Province using a multisource
dataset (1997–2025).
Results demonstrate state-of-the-art performance. A weighted LSTMNeuralProphet
ensemble achieved an AUC-ROC of 0.999, an F1-Score of 0.987, andbalanced accuracy of 0.989. Classical ensembles attained a peak AUC of 0.9981 with
precision and recall exceeding 99.7%. The fuzzy-logic subsystem showed a field deviation
of ±4.8%, identifying temperature as a primary driver (55% influence). Critical
indicators include the Mediterranean Fire Danger Index (MFDI) and Keetch-Byram
Drought Index (KBDI). The IoT infrastructure ensured data integrity with 100% duplicate
elimination, while edge processing reduced latency by 40%.
This thesis establishes a foundational framework for proactive wildfire resilience,
providing a novel Environmental Intelligence methodology, a validated architectural
blueprint, an interpretable hierarchy of context-specific drivers, and a practical
decision-support tool. This work marks a paradigm shift from reactive firefighting
to precision risk management, enhancing forest preservation, public safety, and resource
allocation in fire-prone regions.