Geospatial machine learning for flood risk assessment in contrasting physiographic environments

One of the biggest hydrological hazards in Sub-Saharan Africa is flooding, which is exacerbated by increasing rainfall, inadequate drainage systems, and growing urbanization. In Nigeria, fragmented datasets and inadequate methodological integration continue to limit the ability to map flood susceptibility in a spatially detailed manner. This work presents a hybrid framework that creates interpretable and highly accurate flood susceptibility models for two physiographically distinct regions: Ile-Ife (inland uplands) and Ilaje (coastal lowlands) by combining the Analytic Hierarchy Process (AHP) with the Random Forest (RF) classifier. For Ilaje and Ile-Ife, a total of 43,825 and 8,632 spatial sample points were produced. In order to create a composite Flood Susceptibility Index (FSI), four flood-related predictors elevation, slope, rainfall, and distance to river were normalized and weighted using AHP. To train RF models for each region, the FSI was reclassified into three risk categories. F1-scores, precision, recall, and confusion matrices were used to assess the model’s performance. According to the results, Ilaje and Ile-Ife had classification accuracy rates of 98% and 97%, respectively. In both areas, rainfall and river proximity were the most important predictors, whereas the complexity of the terrain affected the patterns of susceptibility. The AHP-RF framework proved to be highly transparent and dependable, providing a scalable flood risk zoning tool, especially in settings with limited data. This work promotes interpretable geospatial modeling for disaster risk reduction by combining machine learning and expert judgment. The results provide a replicable model for climate adaptation in flood-prone areas of Sub-Saharan Africa and support the incorporation of physiographically informed flood planning into policy frameworks.