This study developed a novel combinatorial approach for flood susceptibility modeling in urban areas with limited historical flood inventory data. The methodology was based on the application of the Analytic Hierarchy Process-Gaussian (AHP-G) for the generation of training samples, which were subsequently used in machine learning algorithms. The study area was the city of Belém, located in the eastern Brazilian Amazon. A spatial database was constructed using a range of geographic conditioning factors, including elevation, slope, aspect, stream power index, topographic wetness index (TWI), height above the nearest drainage (HAND), distance to drainage channels, and flow accumulation area. The Random Forest (RF), Support Vector Machine (SVM), and Generalized Linear Model (GLM) algorithms were applied to produce flood susceptibility maps. Model validation was conducted using in-situ flood occurrence data collected between 2010 and 2023. The reliability of the models was assessed using various statistical performance metrics, with particular emphasis on the area under the receiver operating characteristic curve (AUC-ROC). The results demonstrated that the Random Forest (RF) algorithm achieved the highest predictive accuracy, with an AUC exceeding 90%. The most influential factors in the modeling process were elevation, HAND, soil characteristics, distance to drainage channels, and precipitation. Overall, the study provides robust and high-quality results that can support decision-making processes regarding future flood risk management and urban planning.