Design of a Predictive Model Using AI and GIS for the Management and Optimization of Resources in Forest Fires in Eastern Michoacán

This study describes a design and validation proposal for an artificial intelligence–based predictive model to estimate the area affected by forest fires and support the optimisation of operational resource use in the eastern region of Michoacán, Mexico. Using a dataset comprising 930 historical records (2015–2024), three supervised learning algorithms were evaluated: Multilayer Perceptron (MLP), Random Forest (RF), and XGBoost. The MLP model, optimised using L2 regularisation, Dropout, and cross-validation, achieved the highest performance, with a coefficient of determination (R²) of 0.76, compared with RF (0.57), XGBoost (0.51), and a weighted average ensemble model (0.60). The results were integrated into interactive cartographic platforms using QGIS and Leaflet JS, enabling the visualisation of areas with higher fire susceptibility and the prioritisation of interventions. A replicable tool with low computational requirements is put forward as suitable for institutional contexts with limited resources. Smart citations: https://scite.ai/reports/10.61467/2007.1558.2026.v17i1.1181Dimensions.Open Alex.