<p>This study presents a predictive modeling framework for estimating churning losses in helical gears by integrating simulation and experimental data through a multi-fidelity Gaussian process regression (MFGPR) approach. High-fidelity experiments provide accurate insights but are expensive, whereas low-fidelity simulations are efficient yet less reliable. By combining both data sources, the MFGPR model captures underlying trends from simulations and corrects them using experimental data, improving prediction accuracy. Smoothed particle hydrodynamics (SPH) simulations and a customized experimental setup were used to generate low- and high-fidelity datasets. The performance of each model was compared in terms of <i>R</i>². The results demonstrate that the MFGPR model not only significantly outperforms single-fidelity models but also enables efficient and accurate prediction of churning loss across varying gear geometries and operating conditions. Sobol’ sensitivity analysis identified oil level as the most influential factor. The proposed framework provides an efficient and reliable tool for early-stage gear design optimization.</p>