Accurate and highly generalizable heat transfer correlations are essential for the optimal design of supercritical heat exchangers and the efficient and safe operation of thermal systems. However

an extensive evaluation based on experimental datasets of water

carbon dioxide (CO2)

and R134a flowing in horizontal tubes reveals that existing correlations exhibit significant errors in cross-fluid predictions and fail to capture the common heat transfer characteristics among different supercritical fluids. To improve prediction performance

this study employs a data-driven approach combining the Buckingham Pi theorem and the active subspace method to identify the dominant dimensionless group governing supercritical heat transfer in horizontal tubes. The normalized eigenvalue of the identified dimensionless group reach 0.997

effectively capturing the key characteristics of supercritical heat transfer. Based on the dominant dimensionless group

a new heat transfer correlation is developed

demonstrating excellent cross-fluid predictive capability. For water

carbon dioxide

and R134a

the new correlation achieves a mean absolute relative error of 0.05

with 98.67% of data points having relative errors within 20%. Furthermore

validation using R1234yf demonstrates superior generalization capability compared to existing models. These findings indicate that the identified dominant dimensionless group effectively captures the shared heat transfer behavior of various supercritical fluids

offering a new perspective for developing universally applicable supercritical heat transfer correlations.