A Comparative Analysis of Time-Series Models of ARIMA and Prophet IoT-Based Flood Forecasting in Sungai Melaka

Flood prediction is essential for mitigating disasters, especially in low-lying areas. This study presents an IoT-driven flood forecasting system that utilizes ARIMA and Prophet models to predict water levels in Sungai Melaka, Malaysia. Sensor data collected from an IoT-based flood observatory system was used to train and evaluate both models. Performance analysis based on RMSE and MAPE revealed that while ARIMA captures short-term trends, Prophet outperforms it with a lower MAPE of 6% and RMSE of 5, demonstrating superior accuracy and adaptability. Prophet's advantage lies in its robust seasonality handling, flexible trend adjustments, and ability to incorporate external regressors, making it more effective for real-time flood monitoring. The study also highlights Prophet’s limitations in capturing abrupt water level spikes, suggesting that integrating environmental factors such as rainfall intensity and upstream discharge could enhance predictive accuracy. The findings contribute to the development of AI-driven flood warning systems, supporting urban disaster management strategies.