<p>Urban flooding has emerged as a recurrent menace, posing severe challenges to urban infrastructure, livelihoods, and sustainable urban growth. Developing accurate models for flood susceptibility mapping is imperative for informed hazard mitigation and for strengthening urban resilience. This study develops, compares, and applies six machine learning (ML) and three deep learning (DL) models for urban flood susceptibility analysis of the National Capital Territory (NCT) Delhi, India. Fifteen flood-influencing factors were used in this study. Gradient boosted decision tree (GBDT) yielded the most accurate urban flood risk map, with a receiver operating characteristic-area under the curve (ROC-AUC) of 93.27%, an accuracy of 87.92%, and a precision score of 89.46%. The tree-based models, including GBDT, random forest (RF), and decision tree (DT), are inherently programmed to provide feature importance insights. These models efficiently decipher the complex relationships between urban floods and their influencing factors. These indicate that LULC is the most flood-influencing factor, followed by the MFI, GLG, and DNS. Finally, flood susceptibility assessment of different constituencies in NCT Delhi was predicted using the GBDT model, and it was found that forty-three out of seventy constituencies exhibit higher flooding probabilities, posing elevated risks for the residents and public infrastructure of NCT Delhi. These findings help the government prioritize constituencies for developing and implementing flood mitigation plans.</p>