Flooding stands as one of the most destructive hydro-climatic hazards, inflicting substantial annual losses on human communities and economic infrastructure. Identifying flood-prone zones and creating sensitivity maps are crucial for effective risk management and natural resource planning. This study focused on assessing the performance of three machine learning algorithms—logistic regression (LR), support vector machine (SVM), and gradient boosting (GB)—for flood risk mapping in the Nekarood watershed of Mazandaran province. Thirteen environmental and hydrological factors, including rainfall, slope, elevation, proximity to waterways, land use, and morphometric indices, were analyzed alongside 152 recorded flood occurrence points. The models were evaluated using metrics such as AUC, overall accuracy, and the kappa coefficient. Results indicated that the GB model achieved the highest performance, with an AUC of 0.896, an overall accuracy of 87%, and a kappa coefficient of 0.84. The SVM model followed with an AUC of 0.872 and an accuracy of 83%, while the LR model, scoring an AUC of 0.853 and an accuracy of 80%, showed the weakest performance among the three. Analysis of variable importance revealed that rainfall, slope, distance from waterways, and elevation are the most significant factors influencing flood occurrence. Consequently, reinforcement learning-based algorithms could serve as effective tools for enhancing predictive accuracy and minimizing uncertainty in flood risk mapping efforts. The findings from this study offer valuable insights for strengthening early warning systems, restricting development in high-risk zones, and reducing both human and economic damages caused by flooding in similar regions.