Hybrid Forest Fires Prediction (HF2P) Strategy Based on Ensemble Classification of Convolutional Neural Networks (CNN) and Decision Tree (DT) models

Hussein, Shaimaa; Saleh, Ahmed

doi:10.21608/njccs.2025.381784.1050

Hybrid Forest Fires Prediction (HF2P) Strategy Based on Ensemble Classification of Convolutional Neural Networks (CNN) and Decision Tree (DT) models

Document Type : Original Article

Authors

shaimaa hussein ¹
Ahmed Saleh ²

¹ Communication Dept. Delta higher institute for engineering and technology.

² mansoura university faculty of engineering

10.21608/njccs.2025.381784.1050

Abstract

Forest fires have become an increasing global concern, causing devastating environmental, economic, and social impacts. Climate change, deforestation, and extreme weather conditions have contributed to the rising frequency and intensity of forest fires, making early detection and prevention crucial. Traditional forest fires prediction methods rely on meteorological data and historical fire patterns, but these approaches often lack the accuracy and speed required for effective disaster management. Deep learning (DL), a subset of artificial intelligence, has emerged as a powerful tool for predicting forest fires with high accuracy. In this paper, Deep Learning (DL) has been incorporated with Decision Trees (DT) to make benefit from the advantages and strengths of both techniques. Convolutional Neural Networks (CNNs) can analyze satellite images to identify high-risk areas, while DTs can assist in making accurate predictions of fire outbreaks. Hence, by leveraging vast amounts of real time data, accurate patterns can be detected that may be overlooked by traditional methods, allowing for more precise and timely predictions. The main contribution of this paper is to introduce a Hybrid Forest Fires Forecasting Strategy (HF2S), which combines evidence from both CNN and DT models. Initially, VGG16 has been employed for extracting features from satellite images, then Gray Wolf Optimization (GWO) is employed to select the most relevant features.

Keywords