Research on Glass Composition Classification Model Based on Random Forest and System Clustering

• Published in: 2023 International Conference on Computer Simulation and Modeling, Information Security (CSMIS) pp. 535 - 541
• Main Authors: Xiong, Shiwei, Song, Minhui, Shi, Tianyu
• Format: Conference Proceeding
• Language: English
• Published: IEEE 15.11.2023
• Subjects: Clustering algorithms; Correlation; elbow rule; Glass; grid search; Lead; mRMR algorithm; Prediction algorithms; Random forest classification; Redundancy; system clustering; Weathering
• DOI: 10.1109/CSMIS60634.2023.00102

In order to study the classification rules of high-potassium glass and lead-barium glass in ancient glass products, this paper selects the data of question C of the 2022 National College Students ' Mathematical Modeling Contest of the Higher Education Council Cup for analysis. In this paper, the random forest classification model in machine learning is established, and the ancient glass relics are classified by using grid search to adjust the optimal parameters. At the same time, the importance value and threshold of weathering and chemical composition are calculated, and finally eight important indexes are selected. According to the eight important indicators and combined with the system clustering method, the high-potassium glass and the lead-barium glass are divided into three and four sub-categories respectively. At the same time, the polymerization coefficient line chart is drawn according to the elbow rule to verify the rationality of the clustering number. By constructing the control sample and fluctuating the chemical composition content by 10 %, it is found that the clustering model of high-potassium glass is more sensitive to the silica content, while the clustering model of lead-barium glass has strong robustness. Finally, the model is applied to predict the type and sub-category of the unknown type of glass relics according to the chemical composition of the glass relics, and the maximum correlation minimum redundancy algorithm is used to find that the binary classification results are sensitive to the four chemical components of weathering, magnesium oxide, lead oxide and barium oxide.