Composition analysis and identification of ancient glass objects based on LightGBM

In this paper, for a batch of existing data related to ancient glass objects in China, by analyzing the relationship between the basic information of cultural relics and the information of chemical composition detected by cultural relics, a prediction model that can predict the chemical composition data before weathering by the chemical composition data after weathering is established, a classification model based on the existing data is built to simulate the original classification law, and the model is used to identify the type of unclassified cultural relics, and then by The importance of the feature selects barium oxide as a criterion for subcategory classification. The proportion of chemical composition of the classified glass artifacts was determined by feature engineering to determine whether the 0 in the detection data was filled in artificially, and it was used as a new feature to reduce the impact of extreme data 0 on the model. The data were then used for model training to obtain the LightGBM classification model. The mean value of the chemical components with the highest feature importance in the model was selected as the criterion for subclass classification, and the values in each class were divided into 2 subclasses. The chemical components with the top 3 feature importance were subjected to sensitivity analysis, and sensitivity judgments were made by the indicators of the model. After first performing feature engineering on the data information of the unknown category of artifacts, the type identification output results were performed using the classification model built in Problem 2. The top 3 chemical components of importance in the model were selected to let their values fluctuate up and down by 5%, and the model indicator curves were plotted for sensitivity analysis.