A logistic regression-based model for classification and evaluation of chemical composition of silicate glass products

Abstract

Glass is one of the earliest human-made materials, and it has played an irreplaceable role in the development of human civilization. In ancient China, glass was an important commodity on the Silk Road, and it became a valuable witness to early trade and cultural exchanges between China and the West. Due to manufacturing processes, ancient glass is easily affected by burial environments and undergoes weathering. During the weathering process, there is a significant exchange of elements between the glass and the environment, causing changes in the proportions of the glass's chemical composition, which can affect the correct determination of the glass type. In this paper, we studied multiple sets of weathered and unweathered silicate glasses, extracted data on the chemical composition, color, and patterns of the sampling points, and first found a significant difference between the six chemical composition elements in the unweathered parts for the two silicate glass categories. Then, we introduced logistic coefficients and established a chemical composition evaluation model based on logistic regression to obtain the classification patterns of the two glass types. To further subdivide the glass into subcategories, we selected the top five chemical composition contents with the highest standard deviation for clustering under each of the two glass types and divided them into three subcategories. We obtained specific partitioning methods and results and analyzed the rationality of the results.