<p>Due to the increasing frequency of climate extremes, predicting hydrological hazards such as snow avalanches has become a critical priority in environmental protection and risk management. This study presents a comprehensive approach to mapping snow avalanche susceptibility in the Šar Mountains National Park (southern Serbia), where avalanches pose significant risks to people, infrastructure, and biodiversity. Using a hybrid methodology that integrates machine learning, deep learning, and fuzzy logic within a GIS framework, susceptibility maps were developed based on nine conditioning factors: normalized difference snow index (NDSI), slope, land use, elevation, plan and profile curvature, aspect, winter precipitation, and winter air temperature. A total of nine models were applied: Logistic Regression (LR), Support Vector Machines (L-SVM, G-SVM), Decision Tree (DT), Random Forest (RF), LightGBM, XGBoost, Multi-Layer Perceptron (MLP), and Fuzzy AHP (FAHP). Avalanche inventory data were used to train and validate the models using a double cross-validation scheme. All models showed strong predictive performance (AUC &gt; 90%), with RF achieving the highest AUC (99.9%) and DT the lowest (93%). Key predictive features were slope, elevation, land use, and NDSI. High-risk zones were found near mountain peaks and ridges, and vulnerable settlements included Restelica, Brod, and the Brezovica ski resort. The area classified as ‘’very high susceptibility’’ ranged from 4.24% (XGBoost) to 24.73% (DT). This is the first study to apply MLP independently for avalanche susceptibility and among the first to integrate diverse AI-based models with fuzzy logic. The results provide valuable support for land-use planning, civil protection, and park management.</p> Graphical Abstract