Subsampling approach and data-driven models to predict silicate glass melt viscosity

•Glass melt viscosity remains one of the most challenging property to predict.•An innovative approach of data subsampling was developed.•Viscosity prediction appeared as very accurate over a wide range of silicate glass compositions. This study focused on developing a predictive tool for calculating glass melt viscosity between 900 °C and 1500 °C. A database containing approximately 16000 silicate glass compositions was built using both literature data and a proprietary dataset. The approach integrates statistical techniques, including design of experiments, machine learning, and subsampling strategies for model training. Prediction accuracy was found to be highly promising for the various types of silicate glasses studied. The relative error in viscosity prediction at 1200 °C was approximately 20 % for simple SBN compositions, and less than one order of magnitude for more complex compositions.