Geographical and Varietal Authentication of Algerian Extra Virgin Olive Oil via UV –Vis and FTIR Spectroscopy Coupled With Machine Learning

ABSTRACT

This study assesses the effectiveness of Fourier Transform Infrared (FTIR) and UV–Vis spectroscopy for the geographical differentiation of Algerian virgin olive oil. Forty‐five monovarietal samples from 11 production areas were characterized by multiplatform spectroscopic fingerprinting and chemometric modeling. Unsupervised analysis of the full sample set by Principal Component Analysis and Hierarchical Cluster Analysis resolved groupings consistent with altitude and cultivar, with the first three principal components accounting for 94.1% of the variance in the UV–Vis data. Supervised classification was restricted to the five regions represented by at least five replicates ( n  = 28) and validated by 10× repeated stratified fivefold cross‐validation, with all preprocessing refitted within each training fold. Against a no‐information rate of 0.25, Support Vector Machine (0.689 ± 0.182), Linear Discriminant Analysis (0.627 ± 0.150) and Random Forest (0.616 ± 0.192) performed comparably and were not statistically distinguishable, while XGBoost was less accurate (0.473 ± 0.202); Cohen's κ ranged from 0.34 to 0.61. A label‐permutation test confirmed that performance exceeded chance ( p  = 0.002). These results indicate that untargeted spectroscopic fingerprinting is a promising rapid screening approach for verifying the provenance of Algerian olive oil, while highlighting the need for larger, balanced datasets spanning multiple harvest seasons before regulatory application.