Ancient Glass Composition Analysis and Sub-classification Methods Based on Interpretable Multi-model Fusion
Due to the influence of the burial environment,the ancient glass will weather because of its internal elements exchanging with the environmental elements,resulting in changes in the composition proportion,which will have a certain impact on archaeological work.In this paper,we analyzed the chemical compositions of glass artifacts from the perspective of content,using univariate factor analysis,chi-square test,and SHAP-SVC fusion method to analyze the three indicators related to the weathering degree of ancient glass,which were ranked as glass type>decoration>color.Taking the type of glass as a categorical variable and the chemical composition content of glass as a presenting variable,the contents of 14 chemical compositions were visualized and analyzed by data mining,and the statistical laws of the chemical composition content with and without weathering on the glass surface were obtained;that was,when the SiO2 content of high potassium glass was higher than 90%,the weathering phenomenon was likely to happen,and when the SiO2 content of lead-barium glass was lower than 30%,the weathering phenomenon was expected to happen.Then,the weathering prediction model was constructed using the difference in the median content of each component before and after weathering.The coarse classification and sub-classification model of glass was established based on GMM and the decision tree algorithm.It was given that the lead-barium glass subclasses were mainly divided by the content of PbO,SiO2,SrO,BaO,and CaO.In contrast,the high potassium glass subclasses were divided by the content of CaO,Al2O3,and SiO2.
ancient glassglass weatheringweathering predictionglass classificationSHAP-SVCdecision tree
万方数据
维普
