Raman Spectroscopy and One‐Dimensional Convolutional Neural Networks for the Forensic Identification of Red Stamp Inks in Questioned Documents

ABSTRACT The accurate classification of ink samples, which are inherently complex mixtures, is critical for verifying the authenticity of historical artworks and financial documents such as contracts, insurance claims, wills, and tax records. Herein, one‐dimensional convolutional neural network (1D CNN) models were developed for Raman spectral data acquired from red stamp ink pastes. The 1D CNN model trained on the first‐derivative Raman spectra showed an F 1 score of 0.964 for the classification of 15 distinct products. Furthermore, the use of data point attribution via gradient‐weighted class activation mapping enhanced the interpretability of the model, revealing the spectral features that contributed the most to the classification decisions. The practical utility of the proposed model was assessed through the classification of unknown samples. The results indicate that 1D CNNs are promising tools for the identification of red stamp inks and can advance forensic and analytical methodologies in this field.