Accurate prediction of rising bubbles in liquids is essential for optimizing gas-liquid processes in environmental systems, bubble columns, and heat exchangers. Existing empirical correlations are limited in generality, while conventional data-based models are often opaque. To address the challenges, this study combines domain knowledge with data-driven discovery to develop a knowledge-informed symbolic learning (KISL) framework that creates a generalizable, explicit velocity model for rising bubbles. The KISL model is established and validated using over 7000 measurements from 58 independent experimental studies covering 46 distinct gas-liquid combinations. It achieves high accuracy (coefficient of determination, CD = 0.96) and delivers substantial improvements over established correlations, reducing the root mean square error (RMSE) by 23–45% and the mean absolute error (MAE) by 28–48%. Compared with benchmark data-driven models, KISL reduces RMSE by 35–52% and lowers MAE by 40–50% when applied to independent datasets. Beyond predictive gains, the KISL framework enhances scientific understanding by revealing physically consistent parameter interactions and providing mechanistic insight that black-box approaches cannot. The resulting formulation is compact, transparent, and embeddable into process optimization tools and uncertainty-based evaluation frameworks. These findings establish KISL as a reliable physics-data hybrid modeling strategy and a transferable approach for advancing predictive modeling of multiphase processes.