An Explainable Raman Spectral Classification Pipeline via NMF and SHAP: A Case Study of Pen Ink Colors

Ink analysis, especially the analysis of coloring components, plays a crucial role in forensic science, as it provides useful information to validate the authenticity of the questioned documents. Raman spectroscopy offers a nondestructive and time- and cost-effective alternative to forensic investigations, providing unique spectral fingerprints of pen ink samples. There are several attempts to analyze Raman spectra using computational techniques. This work proposes an explainable pipeline for Raman spectral classification to distinguish pen ink colors. Preprocessing steps were applied to improve the spectral signal-to-noise ratio. Data augmentation was conducted to increase the training data. Using the weights of the 48 extracted non-negative matrix factorization (NMF) components as inputs for the Random Forest classifier, the test accuracy of 0.8456 was achieved. Moreover, NMF allows underlying linear compositions of spectra associated with ink color classes to be observed. SHapley Additive exPlanations (SHAP) then enables a quantification of the average impact of each NMF component on the classification. Through this pipeline, the scalar multiplication between NMF components, their weights, and their SHAP values enables a human-intuitive explainability of the model prediction results that captures unique Raman spectral patterns as expected for each ink color class.