Model precision in partial least squares with discriminant analysis: A case study in document forgery through crossing lines

Abstract

AbstractDocument frauds can occur by the misuse of blank documents by persons who are trusted by the signer or by the adulteration of an official document. The identification of forgery documents by crossing lines requires methods with better accuracy and precision in nondestructive ways. In this sense, this work presented a methodology by applying the partial least squares with discriminant analysis (PLS‐DA) chemometric tool to digital images obtained from a smartphone for crossing lines analysis in two situations: documents that were printed and then signed (the correct mode—Situation 1) and documents that were signed and then printed (fraudulent documents—Situation 2), and by employing blue pen inks types ballpoint, rollerball, gel, and felt‐tip. PLS‐DA models presented a correct classification of all pen types in both situations and presenting sensitivity and specificity equal to 1. Robustness, evaluated by change the printer brand, model, and ink application mode, showed no influence for gel pens. This model was validated by precision estimation at levels of repeatability, intermediary, and reproducibility showing comparable results for the three levels considering the reproducibility with an iPhone 7. Precision at the reproducibility level with iPhone Xs presented the lower value and probably was more effective due to the camera system.