Using data-driven models to predict floods in advance is one of the current effective methods and hot researches to reduce urban flood disasters. In order to improve the prediction accuracy, it is necessary to select the appropriate flood hazard factors and the number of training samples to construct the prediction model. In our current research, an artificial neural network (i.e., the back-propagation neural network, BPNN) model was developed to predict the flood depth in the next hour. A case study of the urban flood during six typhoons in Macau of China was conducted to prove the performance of the proposed model. The flood depth was collected as output; after analyzing their correlation to the flood typhoon optimum track, urban weather, tides, geographic height and water depth increment of the submerged area were used as input. As a result, four models trained with different sample numbers were developed for training and testing. The model performances were examined using average absolute error, root mean square error and the coefficient of determination. The results show that in this case study, the 30-min scale model provides reliable predictions and can provide useful decision support for the prevention and mitigation of flood disasters in coastal urban.