To improve the efficiency of isolation design and solve the problem that the theoretical formula of stiffness cannot accurately calculate the actual stiffness, this paper proposes a data-driven Artificial Neural Network-Particle Swarm Optimization (ANN-PSO) hybrid model to predict the key parameters of horizontal shear and vertical compression performance of Lead rubber bearings (LRBs), achieving small deviations and surpassing current design code accuracy. These parameters can characterize the elastoplastic behavior of bearings. In this paper, a procedure for building machine learning prediction models is put forward, consisting of the phases of data preprocessing, hyperparameter fine-tuning, and model verification. The dataset, consisting of 200 sets of vertical compression and horizontal shear test results obtained from LRB, was compiled. For machine learning model development, the dataset was randomly divided into distinct training, testing, and validation sets. The performance of ANN-PSO model was assessed by comparing with Artificial Neural Network (ANN), Random Forest Regression (RFR), and eXtreme Gradient Boosting (XGBoost). To evaluate the accuracy of these models, typical statistical metrics were calculated, including the coefficient of determination (R²), the root mean square error (RMSE), and the mean absolute error (MAE). Simultaneously, based on the experimental results, an evaluation was conducted on ANN-PSO model and the stiffness calculation formula specified in the Chinese rubber isolation bearing standards. Results suggests the ANN-PSO model offers greater accuracy. Predicted key parameters for vertical compression and horizontal shear performance show deviations from experimental values of under 20% and 15%, respectively, significantly outperforming the established theoretical design values in the standards.