Identification of ancient glass products based on K-means composition analysis

Abstract

In order to study the classification laws of glass types, the data were first divided into two categories of weathered and unweathered points, and then k-means cluster analysis was used to subdivide each category of data into two categories. It was found that the artifacts in these two categories corresponded to high potassium glass and lead-barium glass, respectively, indicating that k-means cluster analysis could be used as a classification law for high potassium glass and lead-barium glass. Since there are 14 chemical components in each of the four categories, it is more difficult and complicated to use them as the basis for subcategory classification, so principal component analysis was applied to reduce the dimensionality, and the 14 chemical components were replaced by comprehensive indicators (principal components) filtered by the cumulative contribution of eigenvalues over 80%. Then, the sample glass was classified into 15 classes by applying SPSS software to classify the principal components of each class as variables, respectively, and the samples as one event for clustering. In order to verify whether the classification method established by this model is realistic, the results of the division of each category into classes were analyzed separately using ROC curves for reasonableness and sensitivity in this paper, and the final reasonableness and sensitivity were both good.