This study proposes a methodology for flood risk assessment in urban areas using a Deep Neural Network (DNN) optimized with genetic algorithms (GAs). The research focuses on two case studies: Ibadan, Nigeria, and Metro Manila, Philippines. The DNN model is trained on remote sensing data, including precipitation, latitude, longitude, elevation, topographic features, and drainage patterns. The results show that the optimized DNN model significantly improves flood risk assessment accuracy, achieving 0.98 in Ibadan compared to 0.38 with only location and precipitation data in Manila. By incorporating soil data and reducing the number of classes, the model can predict flood risks more accurately, providing insights for proactive flood mitigation strategies and urban planning. The study highlights the importance of selecting relevant and diverse features to build effective machine learning models for flood risk categorization tasks. The GA-optimized DNN techniques demonstrate high accuracy and efficiency, outperforming traditional physics-based hydraulic models in terms of speed, efficiency, and scalability. However, challenges such as interpretability and the need for high-quality data are discussed, along with directions for future research, including handling higher spatial resolution data and integrating more comprehensive datasets.This study proposes a methodology for flood risk assessment in urban areas using a Deep Neural Network (DNN) optimized with genetic algorithms (GAs). The research focuses on two case studies: Ibadan, Nigeria, and Metro Manila, Philippines. The DNN model is trained on remote sensing data, including precipitation, latitude, longitude, elevation, topographic features, and drainage patterns. The results show that the optimized DNN model significantly improves flood risk assessment accuracy, achieving 0.98 in Ibadan compared to 0.38 with only location and precipitation data in Manila. By incorporating soil data and reducing the number of classes, the model can predict flood risks more accurately, providing insights for proactive flood mitigation strategies and urban planning. The study highlights the importance of selecting relevant and diverse features to build effective machine learning models for flood risk categorization tasks. The GA-optimized DNN techniques demonstrate high accuracy and efficiency, outperforming traditional physics-based hydraulic models in terms of speed, efficiency, and scalability. However, challenges such as interpretability and the need for high-quality data are discussed, along with directions for future research, including handling higher spatial resolution data and integrating more comprehensive datasets.