Authentication of Robusta Coffee Origins by Shortwave NIR Spectroscopy Coupled with Dimensionality Reduction and Neural Networks

Non-destructive techniques such as spectroscopy are widely used to authenticate the geographical origins of food and agricultural products. This study presents an integrated approach using shortwave near-infrared (SWNIR) spectroscopy, dimensionality reduction, and artificial neural networks (ANN) to authenticate Robusta coffee beans from four regions in Indonesia: Temanggung, Toraja, Dampit, and Lampung. Spectral data collected in the 954–1700 nm range were transformed using three linear dimensionality reduction methods—principal component analysis (PCA), partial least squares (PLS), and linear discriminant analysis (LDA). The resulting feature sets were used to train ANN classifiers. PCA, PLS, and LDA score plots demonstrated clear clustering among coffee origins. Results show that the LDA–ANN combination achieved the highest classification accuracy of 100, along with perfect values for precision, recall, specificity, and F1-score. In contrast, PCA–ANN and PLS–ANN reached accuracies of 97.9 and 96.2, respectively. The ROC and AUC analysis further confirmed the superior separability of LDA-based classification, showing no overlap between sample classes. These findings highlight the potential of SWNIR spectroscopy combined with LDA and ANN for rapid, reliable, and non-destructive geographical authentication of Robusta coffee. © 2025, Unique Scientific Publishers. All rights reserved.