Fuzzy clustering algorithms for identification of Exocarpium Citrus Grandis through chromatography

Chromatography has been extensively applied in identification and quality control of Chinese medicines (CMs). However, regular analytical methods are not suitable if labeled patterns or reference patterns are not available. Unsupervised and semi-supervised recognition approaches for chromatographic patterns, namely nonrandomized fuzzy C-Means clustering (FCM) with weighted principal components (NWPC-FCM) and partial supervised FCM with weighted PCs (PSWPC-FCM) are proposed in this work. The basic ideas of the proposed algorithms are as follows: PCs are extracted and weighted according to corresponding variances via principal component analysis to search for more complicated geometry of fuzzy clusters, then nonrandomized methodology and partial supervised clustering with seeds are employed, respectively, in NWPC-FCM and PSWPC-FCM to determine initial cluster centers for reliable cluster results. Satisfactory results were achieved with this method in identification of Exocarpium Citrus Grandis, a genuine herbal medicine of Guangdong Province. The presented algorithms improve cluster effectiveness and reliability significantly compared with standard FCM, PC-FCM, and two widely utilized clustering methods on chromatographic analysis. The research indicates the proposed algorithms exhibit functional applicability and interpretability for pattern recognition in chromatographic fingerprints of CMs in the presence of limited labeling or reference information.