Simultaneously Verifying the Original Region of Green and Roasted Coffee Beans by Stable Isotopes and Elements Combined with Random Forest

Abstract

Simultaneously verifying the original region of green and roasted coffee beans is very important for protecting legal interests of the stakeholder according to the chemical analyzing method. 131 green coffee bean samples are collected from six different original regions and pretreated with three degrees (green, middle, and dark roasted); five stable isotope ratios (δ13C, δ14N, δ18O, δ2H, and δ32S) and twelve elemental contents (Al, Cr, Ni, Zn, Ba, Cu, Na, Mn, Fe, Ca, K, and Mg) of green, middle, and dark roasted coffee bean samples (131×3) were analyzed. Fractionation of stable isotopes and variation of elemental contents were evaluated, only isotope hydrogen (2H) significantly fractionated, and elemental concentrations increased with a certain rate during the roasting process. One-way analysis of variance (ANOVA) was used to compare the stable isotope ratios and elemental concentrations of all coffee bean samples from six different original regions. Random forest (RF) was employed to build a discriminating model for simultaneously verifying the original regions of green and roasted coffee bean samples; this model provided 100% accuracy. Inclusion of this mathematical model for simultaneously verifying the original region of green and roasted coffee beans had powerful distinguishing capability and which will not be influenced by fractionation of hydrogen (2H) and variation of element contents during the roasted process.

Stable isotope and trace element analyses with non‐linear machine‐learning data analysis improved coffee origin classification and marker selection

“…22 As of 2022, only six studies had combined these two techniques. [23][24][25][26][27][28] Nevertheless, these articles mostly only looked at linear chemometrics methods such as PCA and linear discriminant analysis (LDA), also known as canonical discriminant analysis (CDA), 24,25,27 to assess the model performance after integrating both datasets. Exploring more advanced non-linear machine learning chemometrics models on the integrated data is necessary.…”

“…Integrating both techniques is even more successful in origin discrimination 22 . As of 2022, only six studies had combined these two techniques 23–28 . Nevertheless, these articles mostly only looked at linear chemometrics methods such as PCA and linear discriminant analysis (LDA), also known as canonical discriminant analysis (CDA), 24,25,27 to assess the model performance after integrating both datasets.…”