In order to investigate the classification laws of the two types, three machine learning models (decision tree, SVM) were constructed in this paper, and their classification accuracy was 96%, which met the practical requirements. Subsequently, a K-means algorithm was constructed to classify the subclasses, and the high potassium and lead-barium glasses were divided into three subclasses. By descriptive statistics of the differences between the subclasses, the results showed that there existed a better differentiation of the divided subclasses in terms of multiple chemical compositions as well as ornamentation and color, which verified its reasonableness. By setting a perturbation factor (a normally distributed sequence with a mean of 0 and a standard deviation of 3) to test the sensitivity of the classification results, the model classification results did not change after several repetitions of the experiment and showed good robustness.