Flood risk prediction and modeling in Bauchi: Leveraging machine learning models and explainable AI for urban resilience

Abstract

Introduction

Floods are amongst the most destructive weather and climate-related disasters, causing significant loss of life and property globally. Accurate flood risk prediction is crucial for improving disaster resilience and urban planning.

Methods

This study employed artificial intelligence (AI) techniques, specifically Random Forest (RF), XGBoost (XGB), and Support Vector Machine (SVM) models, to predict and model flood risk in Bauchi, Nigeria. Additionally, Explainable AI analysis was utilized to interpret the model outcomes.

Results

The study revealed that high-risk areas have a history of frequent and severe flooding based on RF and XGBoost predictions. Settlement formality, elevation, population, and rainfall were the most influential factors in exacerbating flood risk. The RF model outperformed both XGBoost and SVM, with a precision of 0.857 and ROC-AUC of 0.93, while SVM performed the least, with a precision of 0.757 and ROC-AUC of 0.84.

Conclusion

The findings provide valuable insights for climate action, particularly in flood risk and exposure, and emphasize the role of urban planning and effective disaster risk reduction strategies in enhancing urban resilience.