Portable electrochemical sensing coupled with machine learning for rapid Arabica and Robusta coffee authentication and blend ratio prediction.

In this study, we developed a portable electrochemical sensing platform integrated with machine learning for the rapid identification of coffee species and quantitative prediction of blend ratios. Electrochemical measurements targeting caffeine, chlorogenic acids, trigonelline, and reducing sugars were performed and the roast level was incorporated as an additional analytical feature. The resulting electrochemical fingerprints captured both species-dependent compositional differences and roasting-induced chemical transformations. Using these features, an extreme gradient boosting classification model achieved high accuracy in distinguishing arabica, robusta, and blended coffee samples. For blend-ratio regression, the random-split ANN result (R2 = 0.980 and RMSE = 0.050) was retained as a within-distribution benchmark. Under nested group-aware leave-one-blend-pair-out validation, the ANN ensemble achieved R2 = 0.9121, RMSE = 0.0676, and MAE = 0.0547, with 87.0% of predictions within ±0.10 of the reference blend fraction. External validation using literature-reported compositional datasets provided complementary support of species identification. Compared with conventional analytical methods, the proposed electrochemical-ML framework requires only a small aliquot of coffee extract, and the electrochemical measurement can be completed within 3 min after sample preparation.