Two-dimensional linear discriminant analysis for classification of three-way chemical data.

Area of Science:

• Chemometrics
• Spectroscopy
• Machine Learning

Background:

• Two-dimensional linear discriminant analysis (2D-LDA) is effective for feature extraction in image processing.
• Its application to chemical data, particularly three-way spectral data, remains unexplored.
• This study aims to bridge this gap by evaluating 2D-LDA for chemical classification tasks.

Purpose of the Study:

• To investigate the efficacy of 2D-LDA for classifying three-way spectral data in chemometrics.
• To compare 2D-LDA performance against established methods like U-PLS-DA and LDA with TUCKER-3 or PARAFAC scores.
• To assess 2D-LDA's utility in real-world applications, such as food authentication and quality control.

Main Methods:

• Application of 2D-LDA to simulated and real-world three-way spectral data sets.
• Comparison with classification results obtained from raw spectral data, U-PLS-DA, PARAFAC-LDA, and TUCKER3-LDA.
• Utilized surface autofluorescence spectrometry for Parma ham ageing classification and total synchronous fluorescence spectrometry for vegetable oil feedstock classification.

Main Results:

• All methods achieved 100% correct classification on the simulated data set.
• 2D-LDA demonstrated superior performance on real-world data, achieving 86% for Parma ham and 100% for vegetable oils.
• Compared to other methods, 2D-LDA yielded higher classification rates, particularly for the Parma ham data set.

Conclusions:

• 2D-LDA is a powerful tool for feature extraction and classification of three-way spectral data in chemometrics.
• The algorithm shows significant potential for applications in food analysis and quality control.
• 2D-LDA offers improved classification accuracy compared to traditional methods when applied to complex spectral datasets.