Accurate discharge measurement is essential for effective seepage monitoring in dams, a critical component of dam safety. Weirs are commonly used for this purpose, with the triangular sharpcrested weir as one of the most precise discharge measurement structures. While machine learning (ML) techniques have demonstrated high capabilities in various engineering fields, it remains neglected for discharge predictions in triangular sharp-crested weirs. This study compares the models Support Vector Regression (SVR), Gene Expression Programming (GEP), Artificial Neural Network (ANN), Polynomial Regression (PR) and Regression Tree (RT), with traditional empirical formulas for discharge prediction in such weirs. Among those models the strongest performers are the SVR and GEP. They have scores in R-squared, Root Mean Squared Error and mean absolute Relative Error of 0.9789, 2.59E-03, 0.31% for the SVR and 0.9645, 3.37E-03, 0.43% for the GEP. The SVR performs slightly better, but the GEP gives an explicit formula which facilitates its interpretability over the SVR. Results indicate that the ML models significantly outperform traditional empirical formulas, showing greater capacity for adaptability and accuracy across a wider range of conditions. In fact, the strongest empirical formula for a wide range of notch angles is Greve’s formula which has a RMSE and mean |RE| of 11.1E-03 and 1.44%, more than 3 times worse than the SVR and GEP.