Study On the Composition of Glass Products Based on Soft Voting Model and Grey Correlation Analysis

Abstract

The inspection and identification of ancient glass composition has always been an important research content, this paper on the ancient glass as a medium, for its chemical composition, weathering characteristics and classification rules and other aspects of the study, the establishment of a soft voting integrated learning model to random forest algorithm, SVM algorithms and GBDT algorithms as a base classifier, the use of soft voting on the base classifier for the classification of the results of the integration of the learning, and then to the The chemical composition of unknown artefacts is analysed and classified, and the classification accuracy of soft voting is as high as 96.6%, and the coverage rate of ROC curve reaches 99.1%, which indicates that it is reasonable and effective to solve the problem with the integrated learning model. Finally, by changing the number of decision trees, the classification accuracy rate is found to be unchanged, which verifies that the sensitivity of this model is good. In addition, this paper establishes a grey correlation analysis model to generate grey correlation coefficient matrix heat map for the chemical composition of different categories of glass artifacts, the horizontal coordinate of the heat map is the parent sequence, the vertical coordinate is the subsequence, according to the heat map of the correlation between the chemical composition of different types of glass artifacts to summarize and analyze, and in order to validate the reasonableness of the model, but also its Spearman correlation analysis, the results of the analysis of the two models are similar, so it can be judged that the application of the model is reasonable and effective. The results of the two models are similar, so it can be judged that the results of the grey correlation analysis model applied are reasonable. Finally, a detailed comparison of the chemical composition correlations of the major categories and subcategories is carried out, which is a differential analysis of the correlations of different categories of chemical compositions.