Hemicelluloses are amorphous polymers of various sugar molecules and have been widely utilized in bioenergy, mining, and textile. Through hydrolysis, hemicellulose is transformed into sugar oligomers and monomers. In this paper, we build a global optimal decision tree (GODT) model on an extensive hemicellulose hydrolysis dataset containing 1955 experimental data points from 71 published papers from 1985 to 2019. The GODT model is trained to predict xylose yield from hardwood hemicellulose hydrolysis in batch reactors. Compared with the heuristic method, our global optimal algorithm can obtain an average absolute improvement of 1.54% in test accuracy. Moreover, we demonstrate that the reasoning procedure of predictions is easy to comprehend by human decision-makers, thus contributing an explainable model for biorefining.