<p>Unplanned Urbanisation Drives Urban Flood Susceptibility in Kinshasa</p>

Urban sustainability in sub-Saharan African megacities is now tightly co-dependent on flood-risk governance, yet high-resolution, explainable susceptibility maps remain scarce for the cities most exposed to compound rainfall–urbanisation pressure. We present a reproducible, open-data decision-support framework for Kinshasa (Democratic Republic of the Congo, >14 million inhabitants), built around three complementary machine-learning algorithms Maximum Entropy (MaxEnt), Random Forest (RF) and Boosted Regression Trees (BRT) applied to 112 curated flood occurrences (2015-2025) and twelve environmental and anthropogenic conditioning factors derived from Sentinel-2, SRTM, ESA WorldCover, GHS-BUILT and WorldPop. Models were trained and evaluated using five-fold spatially stratified cross-validation, with an environmentally-constrained pseudo-absence strategy (drawn from steep-slope, high-elevation zones without reported flood events), eight classification metrics at model-specific optimal thresholds (True Skill Statistic), and interpretation through Shapley Additive exPlanations (SHAP). All three algorithms converged on the dominance of anthropogenic and land-surface drivers built-up surface, population density, NDVI and land cover over purely topographic variables (AUC 0.977-0.986; TSS 0.949-0.959). High-susceptibility classes concentrated in eight densely urbanised western communes, consistent with documented flood records. Translating these findings into governance action, we propose a five-lever policy-action matrix linking each dominant driver to specific land-use, drainage, waste-management and green-infrastructure interventions, mapped to responsible actors and to Sustainable Development Goals 11 (Sustainable Cities and Communities) and 13 (Climate Action). The framework is fully transferable to other data-scarce African megacities and offers a concrete, explainable tool for sustainable urban flood adaptation.