Abstract
BACKGROUND: Plant-based milk alternatives (PBMA) are increasingly popular due to rising lactose intolerance and environmental concerns over traditional dairy products. However, limited efforts have been made to develop rapid authentication methods to verify their biological origin. OBJECTIVE: In this study, we developed a rapid, on-site analytical method for the authentication and identification of PBMA made by six different plant species using a portable Raman spectrometer coupled with machine learning. METHODS: Unprocessed PBMA (i.e., blended raw nut/grain) and processed PBMA that mimic the industrial processing procedures (i.e., filtration and pasteurization) were prepared in the laboratory and subjected to Raman spectral collection without any sample preparation. Three machine learning algorithms (i.e., k-nearest neighbor [KNN], support vector machine [SVM], and random forest [RF]) were tested and compared. RESULTS: RF achieved the best performance in recognizing the plant sources for the unprocessed PBMA, with accuracies of 96.88 and 95.83% in the cross-validation and test set prediction, respectively. Due to small sample size and risk of overfitting, classification models for the biological origin of processed PBMA were constructed by combining Raman spectra of the unprocessed and processed samples. Again, RF models achieved the highest accuracy in identifying the species, i.e., 94.27% in cross-validation and 94.44% in prediction. CONCLUSIONS: These results indicated that the portable Raman spectrometer captured the chemical fingerprints that can effectively identify the plant species of different PBMA. Using this nondestructive Raman spectroscopic-based method, the overall analysis from sample to answer was completed within 5 min, providing inspection laboratories a rapid and reliable screening tool to ensure the authenticity of the biological origin of PBMA. HIGHLIGHTS: This study presents a novel method for rapid and nondestructive identification of the plant sources of PBMA (both unprocessed and processed) based on the Raman spectroscopic technique and machine learning algorithms.