Composition Analysis of Glass Artefacts Based on Spearman's Coefficient and Fuzzy Correlation

Abstract

The identification and analysis of the composition of glass artefacts can support the related fields in exploring the history of glass and its processes, as well as the development and progress of glass composition. This paper identifies and analyses the composition of glass artefacts by collecting relevant data and information on glass artefacts through the Spearman correlation coefficient method and fuzzy correlation. In terms of data pre-processing, data with component ratios that did not add up to between 85% and 105% were removed, and then the degree of weathering on the surface of the glass artefacts was initially determined to be closely related to the glass type by applying a chi-square test, and then the highest correlation between the weathered surface of the glass and the glass type was determined by excluding possible interference caused by ornamentation and colour through partial correlation analysis. Fuzzy correlation analysis was then applied to remove abnormal data and calculate the mean values of each chemical composition for different types of glass at different degrees of weathering to discover statistical patterns. Finally, the mean values calculated by the fuzzy correlation analysis were summed using the difference to obtain the predicted results. Six chemical components were also screened out by the Spearman correlation coefficient method, and then the difference quotients of the same components were calculated by averaging these six components for different types of glass respectively. Finally, it was found that in addition to the three elements of potassium, lead and barium, calcium and strontium also have a greater influence on glass classification. The sample data were then clustered by the hierarchical clustering method based on the correlation coefficient method of calculating distances, and the optimal number of clusters was found by calculating the clustering contour coefficients. Then, the data of the clustered subclasses were analysed and subclasses such as soda-lime glass, sulphide glass and aluminosilicate glass were classified. Finally, the clustering clarity values were defined using the entropy weighting method to measure the clustering results, and the Held mean values at different p-values were adjusted for sensitivity analysis of the clustering clarity values.