The data-driving method for glass composition design has become a desirable way to create novel glass materials, thanks to the ongoing development of machine learning algorithms and the progressive collection of glass materials. In the present work, the machine learning approach based on random forest regression algorithm was used to develop prediction model between the composition and performance of glass materials which contains 56 different oxides. In addition, the interpretability was studied by SHAP analysis. The accurate prediction of linear expansion coefficient, density, and elastic modulus in high-dimensional composition space was realized. The obtained models were utilized to forecast about 1.18 million Si-Al-B-Ca-Mg-Na six-component oxide glass composition swiftly. Four preferred borosilicate glass samples were tested. The results show that linear expansion coefficient of sample ranges from 52.00×10-7℃-1 to 58.00×10-7℃-1, the density ranges from 2.34 g/cm3 to 2.39 g/cm3, and the elastic modulus ranges from 67.00 GPa to 74.00 GPa, respectively. These values are consistent with the predicted values and outperform the applicable standards.