A probability model for predicting the shear capacity of corroded RC beams based on hybrid tree-structured hierarchical Bayesian estimation and priority memory BFGS (TEP-L-BFGS) optimized Gaussian process regression (GPR) was proposed aiming at the problem that the objective uncertainty of the input parameters and the subjective uncertainty of the hyperparameter formulation were difficult to be considered in deterministic data-driven methods, resulting in low prediction accuracy. The influence of data dimensionality and scale on prediction accuracy was analyzed. The predictive accuracy and generalization performance of the proposed model were validated by comparing with traditional mechanism-driven methods and other data-driven methods. Results indicate that the proposed method can consider the objective uncertainties related to the geometric parameters and degradation levels of corroded RC beams, as well as the subjective uncertainties associated with the prior distributions of hyperparameters, demonstrating high generalization capability. The dimensionality and scale of input features significantly impact the prediction accuracy of model, making the preprocessing of input features essential. GPR quantifies the uncertainty of predicted values compared to deterministic models, significantly improving the prediction accuracy of shear capacity for corroded RC beams.