Feature-Optimized Random Forest Model for Wildfire Prediction using Weather Information

Objectives: Forest fires present a notable danger to economies and human communities. Accurate forest fire forecasting can support timely reactions, resource allocation, and effective management strategies. Methods: The Recursive Feature Elimination with Cross-Validation approach is used to extract the key features from the dataset that was obtained from Kaggle. This technique combines feature selection and cross-validation to ensure the selected features are well-suited to new data. The Random Forest regressor, which constructs several decision trees and combines their predictions to provide a more accurate and reliable result, will be used to implement the prediction. Findings: Recursive Feature Elimination identified temperature, wind speed, and humidity as the most significant predictors of wildfire occurrence. Less impactful features, such as atmospheric pressure and solar radiation, were eliminated, improving model efficiency without compromising accuracy. The model achieved a Mean Absolute Error (MAE) of 0.15 and a Root Mean Square Error (RMSE) of 0.21, outperforming baseline methods. A high R2 score of 0.92 indicated strong predictive capability and explained variance. Cross-validation confirmed the model's robustness across different data splits, demonstrating consistent performance with minimal overfitting. Early identification of high-risk periods based on weather conditions can enhance wildfire preparedness and resource allocation. Novelty: The novelty lies in integrating Recursive Feature Elimination with Random Forest Regressor to optimize weather-based feature selection, enhancing wildfire prediction accuracy and efficiency for proactive disaster management. Keywords: Wildfire Prediction, Recursive Feature Elimination, Random Forest Regressor, Weather Data, Feature Optimization