Dynamic flood risk prediction in Houston: a multi-model machine learning approach

Abstract

In assessing flood susceptibility in Houston, key geographical parameters such as drainage density, slope, distance from rivers and roads, LULC, and rainfall data were analyzed using machine learning models, including Decision Trees, Random Forest, Gradient Boosting, SVM, and ANN. Performance evaluation using ROC curves and AUC revealed ANN as the most effective model, achieving an AUC of 85.00%, outperforming Decision Trees (78.96%), Random Forest (80.29%), Gradient Boosting (82.16%), and SVM (81.84%). Statistical validation with the Kruskal-Wallis test confirmed significant differences among models (H = 11.35, p = 0.023), with ANN excelling in pairwise comparisons. This study highlights ANN's robustness in flood prediction, providing a crucial tool for urban planning and fostering resilient, sustainable development.

Keywords

• Flood risk prediction
• machine learning techniques
• statistical validation
• urban planning
• geographic information systems (GIS)