A Two-phase Type Identification and Subclass Classification Model for Glass Artifacts

Abstract

This paper focuses on the type identification and subclass classification of glass based on the chemical composition data of glass artifacts. While many papers have been conducted from the perspective of chemistry majors based on direct machine detection of chemical substances, this paper focuses on the analysis of data on the properties and chemical composition of glass artifacts based on mathematical modeling to develop a two-phase model for glass type identification and subclass classification. Since the application scenario of the model is archaeological species of ancient glass, the glass types considered in this paper are high potassium glass and lead-barium glass, which were widely circulated in ancient China and surrounding countries. Phase I was based on binary logistic regression to determine whether the type of glass artifacts belonged to high potassium glass or lead-barium glass. After testing the test set, the accuracy of type discrimination was 94%. The second phase utilizes SPSS based hierarchical clustering algorithm for subclassification. After that, the appropriate number of subclasses to be divided can be derived based on the folded graph of clustering coefficients derived from the elbow rule. Finally, this paper presents an accuracy test of the model through test subsets and suggests that the idea of using mathematical-statistical modeling methods to analyze the chemical composition of substances can be extended to the studies related to the chemical composition analysis of all artifacts.