A Study on the Classification of Subclasses of Glass Artifacts Based on Feature Selection

To explore the subclass types of ancient glass artifacts, first we combined the features provided in the dataset with the data on whether the artifacts were weathered or not, constructed a random forest model, and calculated the relative importance of each chemical component by the VIM (Variable Importance Measures) to give the important factors influencing the classification of major classes. Subsequently, we innovatively extracted the important components by improving the coefficient of variation to give the important factors influencing the classification of subclasses. Then, we construct a K-means clustering model for subclassification and give specific criteria for subclassification. Finally, we conducted the rationality analysis from two perspectives of chemical composition and heritage characteristics; we repeated the experiment to test the sensitivity of the large class division model for the random forest model normally distributed white noise sequence; we introduced Dunn index and contour coefficient for sensitivity analysis of the clustering model.