Comparative assessment of GIS-based multi-criteria decision analysis (AHP) and machine learning (MaxEnt) approaches for wildfire susceptibility modeling in Montenegro

This study provides a national-scale assessment of wildfire susceptibility in Montenegro by comparing two modeling frameworks: a Geographic Information System–based Multi-Criteria Decision Analysis (GIS-MCDA) using fuzzy standardization with the Analytic Hierarchy Process (AHP), and a machine-learning approach based on the Maximum Entropy model (MaxEnt). Historical wildfire occurrences from the Moderate Resolution Imaging Spectroradiometer (MODIS) archive (2001–2024) were analyzed together with initial thirteen geoenvironmental initial causal criteria spanning vegetation, climatic, topographic and anthropogenic factors. After the Variance Inflation Factor (VIF) analysis, eleven causal criteria were retained for further modeling. Results of validation using receiver operating characteristic (ROC) curves showed clear performance differences: GIS-MCDA with fuzzy standardization and AHP weighting achieved low predictive accuracy (area under the curve, AUC = 0.51), whereas MaxEnt performed strongly (AUC = 0.81). These findings highlight the ability of MaxEnt to capture nonlinear relationships and complex interactions among geoenvironmental causal criteria, whereas GIS-MCDA with fuzzy standardization and AHP proved inadequate for reliable wildfire susceptibility assessment in this context. These results confirm the findings of previous studies, while showing an even weaker performance of GIS-MCDA with fuzzy standardization and AHP compared to earlier studies. Importantly, this is the first application of machine learning through MaxEnt for wildfire susceptibility assessment in Montenegro, providing a spatial basis for wildfire management and a foundation for national-scale wildfire risk assessment.