Predictive Modeling of Traditional Korean Paper Characteristics Using Machine Learning Approaches (Part 1): Discriminating Manufacturing Origins with Artificial Neural Networks and Infrared Spectroscopy

This study focuses on machine learning-based approaches in combination with infrared spectroscopy to discriminate the manufacturing origin of Hanji, a traditional Korean paper. Infrared spectra provide useful information about the chemical composition and structural features of Hanji, while principal component analysis and hierarchical clustering extract meaningful patterns related to the manufacturing region. Score plots and hierarchical clustering of the principal components provide enhanced clustering patterns based on manufacturing region by focusing on the spectral region 1800-1200 cm -1 . The clustering patterns are driven by key absorption bands, such as those associated with carboxyl groups, crystalline cellulose, and aromatic rings. In addition, feed-forward neural network classification models that were developed using the spectral data exhibit significant accuracy when classifying the Hanji manufacturing regions. In particular, models utilizing the raw and second derivative spectra in the 1800-1200 cm-1 region exhibit excellent classification performance, indicating the effectiveness of this spectral region for classification purposes. This study demonstrates the effective application of artificial neural networks in conjunction with infrared spectroscopy to characterize and classify Hanji based on its manufacturing region. The results contribute to a better understanding of the unique properties of Hanji and the discovery of new insights from paper cultural artifacts.