Ground coffee is highly susceptible to economically motivated adulteration, necessitating robust analytical meth- ods for authenticity verification. This study investigates the efficacy of Fourier Transform Infrared (FTIR) spec- troscopy combined with pattern recognition techniques for detecting adulteration in ground coffee with barley, chickpea, and date pit powders. Spectral fingerprints of pure and adulterated samples were acquired, pre- processed and analyzed via supervised (Random Forest, K-Nearest Neighbors, Decision Tree) and unsupervised (Principal Component Analysis, Hierarchical Cluster Analysis) learning models. Results demonstrated that spec- tral preprocessing significantly enhanced classification performance, with Random Forest achieving the highest accuracy (94 % training, 83 % testing) among supervised models. Hierarchical Cluster Analysis (HCA) outper- formed all methods, achieving 96.8 % accuracy with perfect classification for pure coffee, date pit, and chickpea samples. This study uniquely provides a comprehensive comparative evaluation of supervised and unsupervised models, highlighting the superior potential of HCA for rapid and label-free screening of coffee adulteration.

Keywords