Groynes are essential hydraulic structures in river engineering and water resources management, employed to regulate, diversion and direct water flow while mitigating riverbank erosion. The design and optimization of these structures necessitate a comprehensive understanding of their hydraulic and geometric behavior across varying flow conditions. This research examines the efficacy of three machine learning models (MLMs)—Artificial Neural Network (ANN), Support Vector Machine (SVM), and Gene Expression Programming (GEP)—in simulating the diverted discharge in a channel within a physical model incorporating L-shaped and T-shaped groynes. A total of 96 experimental data points were utilized, considering three independent input variables: Froude number (Fr), relative length of grpyne (L/B), and angle of the water intake channel (α). The models' performance was assessed using three evaluation metrics: RMSE, MAE, and R². The results, while confirming the potential of all three MLMs for simulating diverted discharge, indicated that the GEP model with a three-gene structure exhibited superior accuracy compared to the other two models for both types of groynes. The values of RMSE, MAE, and R² during the training and testing phases for the L-shaped groyne were 0.9325, 0.9878, 1.2536 and 0.9836, 0.4102, 0.6325, respectively, and for the T-shaped groyne were 0.9025, 1.2534, 1.8502 and 0.9873, 0.3337, 0.4972, respectively. The second and third ranks were attributed to the MLP 3-8-1 and SVM models, respectively.