The radial basis function neural network model is optimized for application scenarios such as uneven data distribution, privacy sensitivity, or lack of direct access to the original data. Firstly, federated learning is applied to the radial basis function neural network, facilitating cross-multicenter distributed collaborative training. Then, genetic algorithms are used to optimize the hyperparameters of the federated radial basis function neural network model. Comparative experiments were conducted using the publicly available National health and nutrition examination survey 2013–2014 (NHANES) age prediction subset dataset and the proprietary HEART dataset. Compared with traditional centralized learning methods, the proposed model demonstrates superior performance, providing new ideas and methods for solving distributed learning and data privacy problems.