The objective of this study is to investigate the applicability of machine learning techniques to the discrimination of pottery production origin. Data on the chemical composition of the body of traditional white porcelain from Korea and China, which ...
The objective of this study is to investigate the applicability of machine learning techniques to the discrimination of pottery production origin. Data on the chemical composition of the body of traditional white porcelain from Korea and China, which have over a thousand years of pottery production history, obtained by ICP analysis, were collected and various machine learning algorithms were applied. Out of the 732 collected data samples, 118 Korean and 28 Chinese samples, which represent 20% of each class, were selected as test samples. The country of production was predicted by the developed model. The accuracy of principal component analysis-linear discriminant analysis was 87.7%, decision tree was 96.6%, K-nearest neighbor was 98.6%, and support vector machine was 99.3%. Rubidium was confirmed to be the most significant factor in discriminating origin by mean decrease in impurity and shapley additive explanations value analysis. These findings were consistent in experiments where the number of Korean and Chinese samples in the training and test data were matched. These results demonstrate the effectiveness of machine learning techniques in identifying the country of origin of Korean and Chinese white porcelain. Machine learning techniques are expected to be more effective in determining the origin of pottery through the interpretation of chemical analysis results in the future.