Glass component analysis based on Fisher's linear discriminant model and random forest classification model

Abstract

The production of ancient glass is the crystallization of the wisdom of the majority of ancient working people in China. The differences in fluxes and the differences in the burial environment during weathering have a great impact on its chemical composition content. In this paper, a relevant classification model is established based on the data of various properties of glass. In this paper, Fisher linear discriminant model and random forest classification model were established according to the content of different chemical components, and Fisher linear discriminant model, which is easier to test for sensitivity analysis, was selected. Next, the clustering results were obtained by systematically clustering high potassium glass and lead-barium glass, and the classification basis was determined by drawing a line graph and then establishing a subclass classification model to classify high potassium glass into low potassium subclasses and high calcium subclasses. Sensitivity analysis was performed by changing the values of key parameters to check the reasonableness and stability of the subclass classification model. A discriminant model was established to initially determine the categories of eight unknown classified glass artifacts. Sensitivity analysis was performed by changing the key variables through the gradient descent method, and the model was obtained to have good sensitivity.