This study used a comparative multi-model approach to investigate flood susceptibility in Uttarakhand’s Tehri Garhwal district. Three different approaches were used: a conventional statistical model Logistic Regression (LR); Random Forest (RF), a machine learning methodology and a hybrid ensemble approach Bagging. In order to map geographical vulnerability, 14 flood-conditioning factors were included into these models. Using the Jenks natural breaks approach for ideal calibration, susceptibility was divided into five levels: Very High, High, Moderate, Low, and Very Low. The spatial flood susceptibility modeling of the region placed 595 mouza (particularly Devprayag and Jakhanidhar) within hazardous high-risk zones, despite the fact that RF and LR data suggest 55% of the district remains relatively safe. The Bagging approach offers a more cautious viewpoint, flagging nearly 17% of the land as highly susceptible to flood. These environmental hazards are driven by a “cascading” effect: extreme storms and melting ice create debris flows, while landslides further congest river channels to exacerbate flooding. Model validation via Receiver Operating Characteristic (ROC) curves demonstrated that the RF model achieved the highest predictive accuracy with an Area Under the Curve (AUC) of 94.48%. The Bagging and Logistic Regression models followed with AUC values of 91.16% and 85.88%, respectively, confirming the superior performance of ensemble machine learning in complex Himalayan terrains. These susceptibility maps serve as essential tools for local authorities to enhance community preparedness against increasing extreme hydrological events.