A new method for the composition analysis and classification of ancient glass products before and after weathering

Abstract

This paper focuses on the composition analysis of ancient glass artifacts. With the help of chemical composition data from different locations of glass artifacts, the chemical compositions of weathered and unweathered points are compared using the chi-square test as well as the visualization of R. We find that the chemical compositions of SiO₂ and K₂O have significant changes before and after weathering. Then the composition of weathered points before unweathering was predicted by using the feature point ratio method and neural network prediction. Further, with the help of tools such as Python, modeling methods such as support vector machine and random forest classification were used to dichotomize high potassium and lead-barium glass and compared with the actual results. Then, by using principal component analysis, clustering heat map, and K-means clustering, the composition-based subclassification of glass was established, and the rationality of the subclassification and classification model was verified with the literature, and finally, the sensitivity and rationality of the classification method was analyzed with the defined sensitivity function. The work of this paper will hopefully contribute to the archaeological identification of the composition of glass products.