Chemical Composition Study of Ancient Glass Products Based on Machine Learning and Clustering Algorithm

Abstract

The measurement of the chemical composition of ancient glass artifacts has far-reaching consequences for archaeological work and chemical research and has aroused scholars' curiosity. The goal of this paper is to understand the differences in chemical composition between different weathering types and different glass types, to discover statistical patterns of intrinsic chemical composition content, and to develop prediction and classification models between chemical composition and glass types and weathering types. To ensure the accuracy of the experimental data, this study evaluates many machine learning models and selects the one that best matches prediction and classification. Cluster analysis is then used to sub-classify the known glass kinds in order to provide the framework for future work on the categorization of glass goods. Neural networks were used with principal component analysis to create a mathematical model to generate a fitted function picture between glass kinds and chemical compositions. As a result, valuable suggestions for future chemical studies and archaeological investigations were made. Finally, glass product chemical composition correlations were examined to investigate the variations in chemical composition correlations across various categories.