A flexible and novel strategy of alternating trilinear decomposition method coupled with two-dimensional linear discriminant analysis for three-way chemical data analysis: Characterization and classification

Abstract

This paper proposes a flexible and novel strategy that alternating trilinear decomposition (ATLD) method combines with two-dimensional linear discriminant analysis (2D-LDA). The developed strategy was applied to three-way chemical data for the characterization and classification of samples. In order to confirm the methodology performances of characterization and classification, a series of simulated three-way data arrays and a real-life EEMs data set involving the characterization and classification of tea samples according to the tea varieties were subjected to ATLD-2DLDA analysis. Further, the obtained results were compared with those obtained by using LDA based on relative concentrations of ATLD (ATLD-LDA), discriminant analysis by N-way partial least square (N-PLS-DA) and 2D-LDA method. For the simulated data sets with respect to different levels of noise and class overlap as well as number of groups, the ATLD-2DLDA always obtains superior classification performances than the ATLD-LDA, 2D-LDA and N-PLS-DA methods. Regarding the real EEMs data set of tea samples, the proposed methodology not only could provide a chemically meaningful model of the data for characterizing the different tea varieties, but also achieved the best correct classification rate (100%) for the test samples, compared with the results of ATLD-LDA (83.9%), 2D-LDA (90.3%) and N-PLS-DA (90.3%). These results demonstrated that the proposed methodology was indeed a feasible and reliable tool for characterization and classification of three-way chemical data arrays in a flexible and accurate manner.

Keywords: Alternating trilinear decomposition; Multi-way classification; Pattern recognition; Tea classification; Two-dimensional linear discriminant analysis.