Decision tree based classification of glass artefact types

Abstract

This paper uses logistic regression, SVM, and decision trees in machine learning to analyse 67 data items from question C of the 2022 GCSE Cup National Student Mathematical Modelling Competition. The data were analysed using systematic clustering, the clustering coefficients were analysed, the number of clusters K was further determined by the "elbow" method, a clear classification pattern between the glasses was obtained, and a grid search method was used to classify the glasses. The results show that the new excavated glass artefacts have been classified using the "elbow" method. The results show that newly excavated glass artefacts are classified by varying PbO content. If the lead oxide (PbO) content is below 5.46%, they are considered high-potassium glasses and vice versa for lead-barium drinks. The content of silica was further used as a boundary to divide the high-potassium glass into two subclasses. Lead-barium glasses are divided into three subclasses using the lead oxide and silica content as the boundary. Using a series of models and algorithms, the classification patterns of different types of glass and their subclasses were clarified, and the results were tested for reasonableness and sensitivity. Such a model can be used to classify newly excavated glass artefacts and can also be modified based on this model for the identification and analysis of other ancient artefacts.