RFE-SVM for discrimination of Pericarpium Citri Reticulatae through chromatography

Chromatography has been widely used in discrimination and quality control for Chinese medicines (CMs). Nevertheless, regular analytical approaches are not applicable if training samples are small while features are large. Support vector machine (SVM) with recursive feature elimination algorithm (RFE-SVM) is presented in this study for discrimination of Pericarpium Citri Reticulatae through small chromatographic samples. The primary strategies of the proposed method are as follows: firstly, features for analysis are extracted according to the aligned chromatographic fingerprints (CFs), then RFE was designed to filter some irrelevant features. As result, the remaining features were adopted as input features for the proposed approach in identification of Pericarpium Citri Reticulatae, a genuine herbal medicine of Guangdong Province. RFE-SVM classifier achieved a satisfactory prediction performance with an accuracy of 93.8%, following the leave-one-out cross validation procedure. The proposed algorithms improve the accuracy of identification significantly compared with SVM, Fscore-SVM and Hierarchical clustering analysis (HCA) on chromatographic analysis. The study demonstrates the presented method shows interpretability and applicability for pattern recognition in chromatographic fingerprints of CMs on the occasion of limited training samples.