Cavitation in ogee spillways is one of the most critical factors threatening the safety and reducing the service life of hydraulic structures. In this study, a deep learning model based on a multilayer perceptron (Deep MLP) was developed to predict different levels of cavitation damage on the Golvandak Dam spillway. To achieve optimal performance, the influence of key parameters such as data splitting strategy, training algorithm, activation function, and learning rate was systematically investigated. The results indicated that a data split of 70-15-15% for training, validation, and test, respectively, combined with the Trainlm algorithm, Tansig activation function in hidden layers, and a learning rate of 0.001, provided the highest accuracy. The final model with three hidden layers (10, 8, and 5 neurons) was able to simulate the nonlinear relationships governing cavitation with high precision, achieving correlation coefficients of 0.998 in training and 0.996 in testing, with RMSE values of 0.0158 and 0.028, respectively. The prediction results revealed that damage severity increases along the spillway and reaches its maximum level from station 90 m onward. Comparison with the Support Vector Machine (SVM) model demonstrated the significant superiority of the deep learning model in both training and testing phases. The developed model can serve as an efficient tool for identifying critical zones, prioritizing maintenance actions, and enhancing the safety of hydraulic structures.