Research on glass relics based on machine learning

The degree of weathering, chemical composition and classification of glass relics are of great significance to the research of archeologists. In this paper, the classification rules of high-potassium and lead-barium glass are constructed using the decision tree algorithm, and then the data set is determined according to the forecast demand. The problem of predicting the type of glass relics is defined as a general binary problem, which is trained by a classical neural network, multi-layer perceptron (MLP). Due to the small amount of data, the parameters were optimized by using the five-fold cross validation, and the sensitivity test was carried out. It was found that the overall robustness of the model was very strong, and the parameters had little influence on the results. After finding the optimal parameters, the prediction of MLP is highly matched with the prediction results of the decision tree, which proves that the model is reasonable.