This study explores and compares the predictive capabilities of various ensemble algorithms, including SVM, KNN, RF, XGBoost, ANN, DT, and LR, for assessing flood susceptibility (FS) in the Houz plain of the Moroccan High Atlas. The inventory map of past flooding was prepared using binary data from 2012 events, where “1” indicates a flood-prone area and “0” a non-flood-prone or extremely low area, with 762 indicating flood-prone areas. 15 different categorical factors were determined and selected based on importance and multicollinearity tests, including slope, elevation, Normalized Difference Vegetation Index, Terrain Ruggedness Index, Stream Power Index, Land Use and Land Cover, curvature plane, curvature profile, aspect, flow accumulation, Topographic Position Index, soil type, Hydrologic Soil Group, distance from river and rainfall. Predicted FS maps for the Tensift watershed show that, only 10.75% of the mean surface area was predicted as very high risk, and 19% and 38% were estimated as low and very low risk, respectively. Similarly, the Haouz plain, exhibited an average surface area of 21.76% for very-high-risk zones, and 18.88% and 18.18% for low- and very-low-risk zones respectively. The applied algorithms met validation standards, with an average area under the curve of 0.93 and 0.91 for the learning and validation stages, respectively. Model performance analysis identified the XGBoost model as the best algorithm for flood zone mapping. This study provides effective decision-support tools for land-use planning and flood risk reduction, across globe at semi-arid regions.