Manila City experiences recurrent flooding driven by its low-lying topography, high population density, and rapid urbanization. Traditional hydrodynamic models, while accurate, are computationally expensive and unsuitable for rapid scenario evaluation. This study proposes a structured, scenario-based flood susceptibility mapping framework using supervised machine learning trained on synthetic hydrodynamic simulations to address these limitations. Synthetic rainfall hyetographs each representing rainfall intensities at 5-minute intervals over a 2-hour duration were used to simulate flood events through two-dimensional unsteady flow modeling in HEC-RAS. The resulting maximum flood extent maps serve as ground truth data for model training. Input features consist of digital elevation model (DEM), soil type, land use, and the rainfall hyetograph vector, all preprocessed into spatially aligned raster datasets. Machine learning classifiers including Support Vector Machine (SVM), Random Forest (RF), and Extreme Gradient Boosting (XGBoost) were trained to identify flooded areas at the pixel level and feature vectors were constructed by combining spatial characteristics with rainfall inputs. The trained models accept new input conditions comprising DEM, soil type, land use, and a defined rainfall hyetograph and produce a corresponding maximum flood susceptibility map. This capability enables flood susceptibility predictions for a wide range of hypothetical rainfall events without the need to rerun simulations wherein resulting maps can inform land-use planning, infrastructure design, evacuation planning, and disaster risk management in flood-prone urban environments such as Manila City.