Research on the Prediction of Ancient Glass Artifact Types based on XGBoost and Random Forest Algorithm

Abstract

Ancient Chinese glass is susceptible to weathering by environmental influences, and its chemical composition has changed considerably thus affecting the category judgment. The classification of ancient glass is mainly divided into two types: high potassium and lead-barium. In this paper, in order to determine the type of ancient glass according to its chemical composition, the multicollinearity phenomenon of many chemical compositions is reduced by using LDA model and then the parameters of XGBoost and RF models are optimized by using GA algorithm and the five-fold cross-validation of existing data sets, and then the weights of XGBoost and RF are optimized by GA algorithm, and finally the GA - XGBoost - RF dual optimization model. The predicted categories can be derived by bringing the chemical composition content of unknown categories of artifacts into the well-trained model, which provides more accurate and faster categorization guidance for ancient glass artifact types to a certain extent.