Hair dyeing is a widespread practice with potential forensic value in individual identification, yet most analytical approaches are destructive, time-intensive, or lack sensitivity for trace residues. Surface-enhanced Raman spectroscopy (SERS) offers a rapid, nondestructive, and highly sensitive alternative. We introduce DyeSPY, the first forensic SERS and machine learning platform for identifying oxidative and nonoxidative hair dye colorants and predicting their perceptual colors. Using spectra from 44 pure colorants, laboratory-prepared mixtures, and 60 commercial dye products applied to hair, we developed a three-phase classification pipeline. Phase I distinguished oxidative from nonoxidative dyes with up to 98.6% accuracy on hair using partial least-squares discriminant analysis. Phase II achieved high-fidelity colorant identification: for nonoxidative dyes, synthetic training via linear spectral mixing yielded an F1 score of 0.88 with 85.7% mean subset recall; for oxidative dyes, an artificial neural network attained perfect hair classification (F1 = 1.00) and 98.5% subset recall for dye solutions. Phase III predicted perceptual colors with ≥97.5% accuracy by using cosine similarity. Validation on external data sets confirmed robust performance despite substrate variability. By integrating chemically informed modeling of stable and reactive dye systems, DyeSPY establishes a forensic-grade framework for accurate and interpretable hair dye analysis.
(Publisher abstract provided.)