Subclassification of Chemical Composition of Glass Based on Principal Component Analysis and K-Means Clustering

Abstract

Accurately carrying out the subclassification of the chemical composition of glass products is of great significance for the compositional analysis and identification of archaeological artefact samples. Firstly, this study collected the relevant data of glass composition and divided it into two groups: high potassium glass and lead-barium glass. Combined with the controlled variable method, the analysis concluded that the weathering condition of the surface of high-potassium glass artifacts has a great correlation with the decoration, while the weathering condition of the lead-barium glass artifacts has no correlation with the decoration and colour. The artefacts were divided into four categories according to the glass type and weathering condition, resulting in a law indicating the effect of weathering on chemical composition. Thus, the weathering point is predicted inversely. Then, this paper carries out a significance analysis of the various components of the two types of glass, respectively, and obtains that the F-values of silica, potassium oxide, and barium oxide are 38.28, 30.31, and 26.99, respectively, which are the most significant. Secondly, using the comprehensive determination model, the principal component analysis method was firstly adopted to obtain the greatest influence of CaO on the overall variance. Then, K-means clustering method was used, and through many iterations, it was finally concluded that: for high potassium glass, according to whether the CaO content is greater than 4 as the subclass division method; for lead-barium glass, according to whether the CaO content is greater than 3 as the subclass division method, and listed all the samples of the subclass division method, and finally verified the reasonableness of the subclass division method and the results, and verify that the model has good sensitivity.