Machine Learning Based Flood Hazard Mapping at the Ward Scale for Indian Coastal Cities

Abstract

The Indian coastal cities of Mumbai and Chennai and Kolkata face an intense urban flooding problem because their flood risk has risen from the combination of increasing ocean heights and powerful monsoon storms and haphazard urban growth. The emergency response requires fast and accurate flood prediction which traditional systems based on historical averages and static thresholds cannot deliver. In this paper an AI-powered flood risk assessment framework designed to provide granular, ward-level flood predictions using realworld geospatial and meteorological data. The system proposal generates risk predictions through machine learning models which process structured weather data together with satellite elevation information and municipal flood records. The predictive models consist of three main components which include an XGBoost classifier for flood risk classification and an XGBoost regressor for flood severity prediction and a Random Forest classifier for ward-level vulnerability assessment. The models produced excellent results because they achieved 86.7% classification accuracy and 0.91 R2 score for severity prediction and 88.2% accuracy in ward-level risk identification. The system delivers its results through an interactive web dashboard which provides stakeholders including planners and disaster response teams and the general public with real-time visualizations and risk heatmaps and data export capabilities. The system shows how artificial intelligence technology improves urban flood prediction by producing exact and useful information when it needs to. The system includes built-in scalability which allows developers to implement real-time sensor data collection and mobile notification systems for upcoming development which will help India create climateresistant urban spaces.