Abstract: This study investigates erosion in the elbows of shale gas gathering and transportation systems caused by sand accumulation. To this end, five parameters-derived from sand particle and pipeline characteristics-were selected as input features to predict maximum erosion rates as outputs. The prediction accuracy of different machine learning (ML) models was compared using datasets obtained from Latin hypercube sampling (LHS) and Fluent simulations. The results show that the PSO-SVM model, which incorporates particle swarm optimization (PSO) and support vector machines (SVM), is identified as the optimal model. Based on the test dataset, this model recorded a mean absolute error (MAE) of 4.859 94×10^-5 and a root mean square error (RMSE) of 5.060 3×10^-5, with a coefficient of determination of 0.98. Compared to the experimental results, its relative prediction error was only 14.84%. Factors influencing the outputs of maximum erosion rates, as revealed through Shapley additive explanations (SHAP), are ranked in descending order of contribution: particle mass flow, particle velocity, particle size, pipe diameter, and radius-to-diameter ratio.