Analysis and identification of glass components based on regression models and improved K-means++ clustering algorithms

This paper proposes a machine learning-based method for identifying and analyzing ancient glass artifact types, with the aim of improving the efficiency of studying such objects. The method uses partial least squares regression models and an improved K-means++ clustering algorithm. To predict the chemical composition of weathering detection sites before weathering, the paper constructs a partial least squares regression model based on chi-square tests and analyses of variance. An Adaptive-LASSO regression model was then used to analyze the correlation between the chemical composition of different categories of glass artifacts. Additionally, a random forest classification model was established to analyze the classification patterns of high potassium glass and lead-barium glass, and feature screening of the chemical composition was carried out. A stepwise prediction model based on Bayesian parameter optimization of random forest was then used to analyze the chemical composition and identify the type of glass artifacts of unknown categories. To improve the K-means++ algorithm, the paper establishes a K-means++ clustering model based on weighted distance, which classifies the two types of glass separately. The method determines that for high-potassium glass, the fitting goodness-of-fit coefficients R for SiO2 and K2O curves are 0.92 and 0.92. For lead-barium glass, the fitting goodness-of-fit coefficients R for PbO and BaO are 0.91 and 0.82, both at a high level. The model fitting effect is good, and the optimal clustering number for both types of glass is K=3, with a reasonable model classification.