Excessive water losses due to leakage have become a major concern for national water authorities. In fact, the Tehran Water and Wastewater Company has been forced to implement complete nightly shutdowns in the city’s water distribution networks to conserve water, causing public dissatisfaction and increasing the risk of contaminant intrusion into the system.This study adopts a data-driven hybrid approach, combining numerical modeling of Reservoir 43—located in northeastern Tehran—using Water GEMS software, and training an Artificial Neural Network (ANN) with pressure and flow outputs from the hydraulic model under leak conditions to predict both location and magnitude of leaks.Model performance and accuracy were evaluated based on RMSE, MSE, and R² criteria. Numerical simulations showed that, after applying leak conditions, the average pressure decreased by approximately 0.9 bar, while flow rates in main pipelines increased by around 35%. The results indicated that the probability of detecting leaks ranged between 67% and 88%.Following full training, the integrated model estimated pressure drops and flow increases caused by leaks with very high precision, reducing the prediction error to 4% by the end of the training period. This high accuracy confirms the model’s reliability for rapid and targeted leak identification under real operational conditions.The success of this method depends on the accuracy of hydraulic model calibration and the quality of measured data. In this research, calibration was performed based on engineering knowledge, adjusting pipe roughness, local losses, and minor loss coefficients until the difference between modeled and field data was less than 5%.