Rainfall trends and flood prediction in Chattogram district: a data-driven approach

Abstract

Flooding poses severe socio-economic challenges in Bangladesh, with nearly one-third of the country highly vulnerable to flood-related disasters. Despite its recurring impact, research on flood prediction remains limited. This study addresses this gap by analyzing rainfall trends and developing a data-driven flood prediction framework for the Chattogram district, a region frequently affected by extreme weather events. Analysis of historical rainfall data revealed intensifying monsoon patterns and identified critical flood thresholds, with annual rainfall exceeding 3,000 mm and cumulative rainfall above 2,000 mm during the June–September monsoon season indicating a high risk of flooding. Several machine learning models were evaluated for flood prediction. Among them, the Random Forest (RF) model achieved the best performance, with 92% accuracy and a 93% F1-score. Logistic Regression performed poorly due to its inability to capture complex patterns, while K-Nearest Neighbors and Decision Trees demonstrated moderate predictive performance. Although Gradient Boosting produced consistent results, it was outperformed by the Random Forest model. The findings demonstrate that integrating rainfall threshold analysis with machine learning techniques can significantly improve flood forecasting, early warning systems, and flood risk management in Bangladesh.