FIRE MITIGATION STRATEGIES IN SFERA@UMS BASED ON FIRE PREDICTION USING RANDOM FOREST

Abstract

Forest fires pose significant threats to ecosystems, biodiversity, and community. In a forested area in a university area, conducting early predictions and assessing mitigation strategies are essential for reducing the impact. This study evaluates the effectiveness of the Random Forest (RF) algorithm in predicting forest fire risk within the Sustainable Forest Education and Research Area (SFERA@UMS) and proposes targeted mitigation strategies. The approach integrates machine learning with GIS-based spatial analysis using variables such as elevation, slope, land use classification, and proximity to roads. Results show that the RF model achieved an overall prediction accuracy of 91%, with high-risk zones concentrated in low-elevation areas with steep slopes and near roads. These conditions increase fire susceptibility during higher temperatures and increase of human activity near or in the forest area, which elevate ignition potential. GIS tools were employed to generate a fire risk map, classifying areas into low, moderate, and high-risk categories for better visualization and planning. Furthermore, Unity-based digital twin technology was utilized to simulate fire spread and assess mitigation measures, including firebreaks, hydrant installations, signage, and community-based fire teams. This research demonstrates the potential of fire occurrences and suitable mitigation strategies to increase fire resilience and improve forest fire management.