A chemometric study of chromatograms of tea extracts by correlation optimization warping in conjunction with PCA, support vector machines and random forest data modeling

PCA, support vector machines and random forest machine learning methods were evaluated comparatively on their ability to predict unknown tea samples using models constructed from a predetermined training set, and the best predictions of identity were obtained by using RF.

Abstract

A reverse phase high performance liquid chromatography (HPLC) separation was established for profiling water soluble compounds in extracts from tea. Whole chromatograms were pre-processed by techniques including baseline correction, binning and normalisation. In addition, peak alignment by correction of retention time shifts was performed using correlation optimization warping (COW) producing a correlation score of 0.96. To extract the chemically relevant information from the data, a variety of chemometric approaches were employed. Principle component analysis (PCA) was used to group the tea samples according to their chromatographic differences. Three principal components (PCs) described 78% of the total variance after peak alignment (64% before) and analysis of the score and loading plots provided insight into the main chemical differences between the samples. Finally, PCA, support vector machines (SVMs) and random forest (RF) machine learning methods were evaluated comparatively on their ability to predict unknown tea samples using models constructed from a predetermined training set. The best predictions of identity were obtained by using RF.