An efficient strategy for early sex identification in Litsea cubeba based on portable Raman technology combined with machine learning algorithms

Early sex identification of the dioecious and medicinal spice plant, Litsea cubeba (Lour.) Pers (LC), is necessary for expanding the production and application of LC. We describe a Raman spectroscopy (RS) and surface-enhanced Raman spectroscopy (SERS) method combined with machine learning techniques. Based on the RS and SERS features, we combined principal component analysis and linear discriminant analysis (PCA-LDA) to dimension reduction for the data. And we combined seven machine learning algorithms, including logistic regression (LR), Naive Bayes (NB), decision tree (DT), random forest (RF), k-nearest neighbor (KNN), support vector machines (SVM), and extreme gradient boosting (XGBoost) algorithms to construct a gender prediction model for LC. The results demonstrated that the surface reinforcement treatment with drops of silver sol combined with LR, SVM, and XGBoost models achieved an accuracy of 84.62% in distinguishing male and female leaves. Furthermore, compared to the discrimination effect of scanning the leaf surface only, the gender recognition accuracy of the surface reinforcement treatment increased by 17.31%, 19.24% and 19.24%, respectively. Portable Raman spectroscopy combined with machine learning algorithms can be promoted for use as a tool for early sex identification in most plants, which can be applied to large-scale plant cultivation and breeding programmes.