Rapid discrimination and quantification of kudzu root with its adulterant part using FT-NIR and a machine learning algorithm

Kudzu root (KR) is regarded as a natural health food with high medicinal value, and is commercially available in dietary supplements that are primarily marketed for women's health. The nutrient and functional substances are mainly distributed in the root; and moreover, the appearance and taste of Kudzu stem (KS) are both similar to KR. Thus, the ability to differentiate KR and KS and to detect adulteration of KR with the addition of cheaper KS is important in preventing the possible commercial fraud that adversely affects both the food industry as well as the consumers. With this assumption, the present work had two different goals of developing a rapid, simple and accurate method to distinguish KR from KS, and to test the feasibility of quantifying the adulteration ratio of KR to KS. To achieve these goals, NIR spectroscopy was coupled with 4 different algorithms including the principal component analysis (PCA), partial least squares-discriminant analysis (PLS-DA), partial least squares (PLS) test, and least squares support vector machine (LS-SVM). The results showed that PLS-DA achieved 100 percent classification accuracy for both KR and KS. The LS-SVM model had superior capability than did the PLS model in predicting the adulteration ratio between KR and KS. It has provided excellent predictive performance with a high R-p value of 0.9972 and a low root mean square error of prediction (RMSEP) value of 2.1724. These observations revealed the potential of NIR spectroscopy combined with machine learning methods as effective approach in distinguishing possible adulteration of KR.