Recognition of counterfeit identity documents using FTIR spectroscopy and machine learning approaches.

Document forgery represents a persistent challenge in forensic science, demanding analytical approaches capable of identifying chemical and structural alterations beyond traditional visual or microscopic inspection. This study proposes a non-destructive methodology for differentiating authentic, forged, and counterfeit Brazilian identity documents using Fourier Transform Infrared (FTIR) spectroscopy combined with machine learning algorithms. Spectra from 135 documents issued between 2003 and 2023 were analyzed through Principal Component Analysis (PCA) and Partial Least Squares-Discriminant Analysis (PLS-DA). PCA revealed characteristic absorption bands associated with authentic and forged substrates, notably linked to calcium carbonate (CaCO₃) and carbonyl (C=O) groups, indicating chemical erasure with carboxylic acid-based solvents. The supervised PLS-DA model achieved high predictive performance, with accuracies of 97.6% for the 1983 layout and 100% for the 2018 layout documents, demonstrating its robustness in distinguishing authentic, forged, and counterfeit documents. The proposed methodology enables the molecular-level recognition of fraudulent alterations, providing analytical evidence of chemical strategies used by counterfeiters and document forgers. Beyond authentication, this approach contributes to the advancement of forensic intelligence by providing a framework for identifying chemical patterns and elucidating the modus operandi of criminal activities. The findings emphasize the potential of FTIR-based chemometric models as objective, fast, and non-destructive tools to support forensic document examination and the broader development of intelligence-driven investigations.