<p>Floods are among the natural disasters that cause the most damage worldwide, making it imperative to identify flood prone areas for effective flood hazard management. However, accurate modeling and forecasting of floods is challenging due to their complex nature. This study presents a novel methodology for the flood susceptibility modelling (FSM) in the Uttar Dinajpur district of West Bengal, India using five boosting-based machine learning (ML) models: AdaBoost, CatBoost, Gradient Boosting Machine (GBM), LightGBM, and XGBoost. A total of twenty Flood Influential Factors (FIFs) were initially considered. Following multicollinearity analysis and Recursive Feature Elimination (RFE) using a Random Forest (RF) model, three factors—geology, normalized vegetation index and stream power index—were excluded, resulting in seventeen key FIFs. Hyperparameter optimization was performed using GridSearchCV, and the models were trained on these selected FIFs and a flood inventory map generated from Sentinel-1 SAR Ground Range Detected (GRD) C-band data. The land use land cover (LULC) map, an essential FIF, was derived using four ML classifiers: SVM, RF, DT, and KNN, with RF chosen as the final method due to its highest kappa accuracy. All boosting models demonstrated strong performance, as validated using statistical metrics such as accuracy, recall, specificity, precision, F1-score, Cohen’s kappa, confusion matrix, and ROC-AUC. Among them, XGBoost outperformed the others, achieving the highest testing accuracy (0.9133) and AUC value (0.9649). Consequently, XGBoost was used for block-level FSM in the district. The results highlight the robustness and reliability of boosting-based ML techniques for FSM. The proposed framework can assist policymakers and planners in developing effective flood mitigation and land-use management strategies in similar flood-prone regions.</p> Graphical abstract <p></p>