Chemical composition analysis of ancient glass products based on decision tree

Abstract

Due to the effects of prolonged burial, freshly unearthed ancient glass is often weathered to varying degrees, and it is difficult to identify the type of glass. We introduce machine learning into the composition analysis and type identification of ancient glass products. This objective is to build a reliable ancient glass classification model based on decision trees and two different k-means clustering algorithms. The performance of the decision tree is measured by the ROC curve. The performance of its clustering algorithm was evaluated by the Calinski-Harabasz index. The results show that the area of AUC in the decision tree is 1 and the highest Calinski-Harabasz index of the two clustering algorithms is 71.68. The predictive ability of the model was verified well.