Development of Flood Risk Prediction Model Using Water Level and Rainfall Volume Using Multiple Regression Analysis

Abstract

Flash floods are a very serious natural risks that usually happened in the tropical areas like Malaysia whereby heavy rainfall causes a quick rise in the surface water levels. Traditional predictive models often rely on single variable indicators, which easily leads to slow or inaccurate hazard assessment. The current research developed a flood risk model, the implementation of involving Multiple Linear Regression (MLR) models along with the use of real-time Internet of Things (IoT) data collection. Tipping bucket rainfall sensor and ultrasonic sensor were used to collect sensor data thus send them using LoRaWAN, and Wi-Fi to a central cloud-based MySQL database. The data streams were treated by a Python based analytical pipeline to make predictions of the flood risk index and classification. The results were sent to the final users over Telegram notification bot and a real-time web dashboard named “FloodWatchAI”. The model was validated on twenty consecutive days empirical observations in Panchor, Johor in Malaysia, collaborating with SENA Traffic Sdn. Bhd and obtained an R^2 coefficient of 0.67 on the test set. The system proved to be associated not only with the practically immediate processing of data and spreading of alerts but also with significantly increased chances of the timely evacuation and proper management of the disaster.