Related Experiment Video
Updated: Jul 14, 2026
08:47
Computer Vision-Based Biomass Estimation for Invasive Plants
Published on: February 9, 2024
An intelligent IoT-machine learning framework for wildfire detection and prediction using a hybrid RF-XGB model
Ahmed A Radhi1,2, Abdullahi A Ibrahim3
1Electrical and Computer Engineering Department, Graduate School of Science, Altinbas University, Istanbul, 34217, Turkey. ahmed.a.radhi@nahrainuniv.edu.iq.
Scientific Reports
|May 13, 2026
Summary
This study introduces an intelligent wildfire monitoring system using a hybrid RF-XGB model and IoT sensors for real-time forest fire detection in Turkey. The advanced framework offers improved accuracy and energy efficiency for early wildfire warnings.
Area of Science:
- Environmental Science
- Computer Science
- Engineering
Background:
- Turkey faces significant seasonal wildfire risks, particularly in summer, due to adverse climatic conditions.
- Existing wildfire detection methods lack comprehensive real-time monitoring and prediction capabilities.
- Scholarly attention on Turkish forest fires remains limited despite high susceptibility.
Purpose of the Study:
- To develop an integrated intelligent wildfire monitoring and prediction framework for real-time detection and risk assessment.
- To propose a novel weighted-voting Random Forest-Extreme Gradient Boosting (RF-XGB) hybrid model.
- To implement an Internet of Things (IoT)-based wireless sensor network (WSN) for enhanced data collection and edge-level processing.
Main Methods:
- Developed an adaptive weighted-voting RF-XGB hybrid model, combining Random Forest and Extreme Gradient Boosting.
- Utilized a multi-season Turkish forest fire dataset with environmental sensor data (temperature, humidity, CO).
- Implemented distribution-preserving sampling and stratified k-fold cross-validation to address class imbalance.
- Deployed a lightweight Multiple Logistic Regression (MLR) model on Arduino Nano for edge-level probability estimation.
- Integrated 80 sensor nodes in Zeytinpark using hybrid grid and K-means clustering for system deployment.
Main Results:
- The RF-XGB hybrid model achieved high performance with 0.9631 accuracy, 0.9627 F1-score, and 0.994 ROC-AUC.
- Demonstrated significant improvements over baseline models: 4.6% RI over XGBoost and 5.7% over Random Forest (AUC).
- Achieved over 50% improvement compared to MLR, SVM, and ANN, indicating superior robustness.
- The edge-computing approach with MLR on sensor nodes estimated an 11-month node lifetime.
- The deployed system in Zeytinpark achieved 95.58% coverage with real-time detection and multi-level alerts.
Conclusions:
- The proposed framework offers a robust, energy-efficient, and scalable solution for rapid wildfire detection and forecasting.
- The adaptive hybrid ensemble design and hierarchical edge-cloud intelligence distribution are key contributions.
- Validated real-world deployment demonstrates the practical applicability and effectiveness of the system for wildfire management.