Classification of ancient glassware based on Random Forest

Abstract

As a precious item in the foreign trade along the Silk Road, it is of great significance to analyze the composition and identify the type of glass relics. In this paper, through the training of random forest, the characteristic importance of each index is obtained, and the classification rule is obtained. It is determined that lead oxide (PbO) is the most important index in the classification of glass types. Then, the k value was determined by the elbow principle, and k=5 was selected as the cluster type. After that, k-means clustering was carried out on the high potassium and lead barium glass respectively to get the classification results, which were divided into 5 subclasses. Due to the small amount of data, in the random forest model, the robustness of the model is constantly enhanced through multiple hyperparameter tuning. In the trained random forest model, it is found that the results do not change, which proves the model is reasonable. At the same time, the correlation between the chemical components of the glass cultural relics samples of different categories was analyzed, and the difference of the correlation between the chemical components of different categories was compared.