Flood AI Monitoring & Early Warning System: A Machine Learning and IoT Integrated Approach Using CWC Data Sandipkumar C. Sagare¹, Ahilya Hapate², Prachi Chougule³, Pranoti Athawale⁴, Rakhi

Abstract: Floods rank among the world’s most catastrophic natural disasters, inflicting severe loss of life, destruction of infrastructure, and socioeconomic disruption. Conventional flood monitoring systems rely on manual observation and static threshold-based alerts, which lack real-time intelligence and fail to provide adequate lead time for evacuation or mitigation. This paper presents a comprehensive AI-powered flood prediction and early warning system that integrates machine learning with IoT-based real-time monitoring. Using hourly river water-level telemetry from the Central Water Commission (CWC) of India, a Random Forest Classifier is trained on engineered temporal features including rate of rise, rolling averages, and danger-level proximity. An ESP32 microcontroller paired with an ultrasonic sensor provides live field measurements, which are processed by a Flask/FastAPI backend and visualized on an interactive dashboard with Twilio-based SMS/email alerts. The system achieves 96% overall accuracy and 97% flood-class precision. Critical analysis of the class-imbalance challenge is provided, with a roadmap for improving recall through SMOTE oversampling and deep learning architectures.

Keywords: Flood prediction, Random Forest, IoT, ESP32, CWC, class imbalance, early warning system, machine learning, SMOTE, real-time monitoring