Classification of printing inks in pharmaceutical packages by LIBS and ATR-FTIR

Abstract The identification of counterfeit pharmaceutical packaging is a complex problem that often requires multiple physical or chemical analyses. In this study, LIBS and ATR-FTIR were used as fast and non-invasive methods to classify pharmaceutical paperboard packaging sample into an authentic group or belonging to one of five counterfeit counterpart packaging sources. The collection set consisted of 124 external pharmaceutical packages, originating from 6 authentic sources (n = 12 packages) and 3 to 5 counterfeit sources per product (n = 112 packages). Orthogonal methods, such as LIBS and ATR-FTIR, were especially advantageous for classification purposes because they reveal chemical information of the organic and inorganic components of the package ink. Different ink colors from the logo, text, barcodes, and images were tested; additionally, the paperboard sample substrate was analyzed with ATR-FTIR. After data reduction via Principal Component Analysis, two supervised machine learning classification techniques were used for classification: k-Nearest Neighbors, and Linear Discriminant Analysis. A random 60:40 split of the data was used for training and testing the algorithms. Across all analyzed ink types, correct classification rates above 70% (LIBS) and 85% (ATR-FTIR) were observed. Data fusion of the two complementary methods improved the results, providing correct classifications ranging from 90 to 100% depending on the classifier and the pharmaceutical package. The results demonstrate that the elemental information obtained by LIBS and the chemical composition by ATR-FTIR are complementary and provide practical alternatives for fast screening of counterfeit secondary packaging.