Safeflow: a multi-zone predictive flood monitoring and early warning system

Flooding remains one of the most destructive natural hazards, particularly in urban environments where rapid runoff and limited drainage capacity intensify water accumulation. Existing flood monitoring systems are primarily reactive, relying on delayed data and single-variable measurements, which limits their ability to provide timely and accurate early warnings. This study proposes SafeFlow, an AI-based flood prediction system that integrates Internet of Things (IoT) sensing with multi-factor hydrological modelling to improve prediction accuracy and responsiveness. The objective of this research is to evaluate the effectiveness of SafeFlow in predicting water levels, providing early warnings, and classifying flood risk using multiple environmental inputs. A synthetic dataset of 3000 samples was generated based on physics-informed hydrological relationships. A feedforward Artificial Neural Network (ANN) was developed using six input variables: water level, rate of change, rainfall intensity, soil saturation, forecast rainfall, and flow velocity. The model predicts water level, time-to-danger (TTD), risk classification, and confidence score. Results demonstrate high predictive performance, with an average error of approximately 2.01 mm and 100% classification accuracy under simulated conditions. Analysis shows that flood severity increases non-linearly due to interactions between rainfall and soil saturation, with a critical threshold observed at high saturation levels. Validation using real-world-inspired data from Shah Alam confirms the model’s ability to generalise and capture real flood dynamics. SafeFlow provides accurate, real-time, and location-specific flood predictions, offering strong potential for implementation in smart city infrastructure, disaster management systems, and early warning platforms.