Prediction and sub-classification of glass original components based on statistical learning algorithm

Abstract

To address the problem that it is difficult for people to identify the information of artifacts affected by weathering, this paper takes several groups of weathered and unweathered silicate glass as the research object, and extracts the data of their chemical composition content and surface decoration, thus analyzing the correlation between decoration, glass type and weathering degree using chi-square test and KS test and studying the proportion of chemical composition of weathered glass before weathering. On this basis, L1 regularized logistic regression was used to roughly classify the initial classification, a subcategorization model was established by hierarchical clustering, and the entropy weight method was used to determine the significant indicators and the optimal number of categories for subcategorization. The results show that glass type has a significant effect on whether weathering occurs; the chemical components that have a significant effect on the determination of the type of lead-barium glass are SiO₂, PbO and BaO, and the chemical components that have a significant effect on the determination of the type of high-potassium glass are SiO₂ and K2O; high-potassium glass is divided into three subclasses, and lead-barium glass is divided into three subclasses.