Tea Classification by Near Infrared Spectroscopy with Projection Discriminant Analysis and Gene Expression Programming

Traditional gene expression programming for classification is designed for binary decisions. Herein, projection discriminant analysis for direct multiclass categorization using gene expression programming is described. Gene expression programming was first employed to examine new synthetic variables that were built as nonlinear combinations of the original features. The data were projected on planes spanned by these new synthetic variables and the nearest centroid was employed to classify new samples. A new objective function was formulated to determine optimum synthetic variables. Direct multiclass categorization using a gene expression programming algorithm was used to classify six tea varieties analyzed by near infrared spectroscopy. Compared with traditional gene expression programming, principal component analysis, and linear discriminant analysis, direct multiclass categorization with gene expression programming algorithm was more efficient. Visual inspection of high dimensional data by this approach also facilitated classification and comprehension of data.