[Classification of pharmaceutical tablet with canonical variates analysis method in spectra analysis]

Area of Science:

• Multivariate data analysis
• Chemometrics
• Statistical pattern recognition

Context:

• Spectral data often presents challenges like multicollinearity and high dimensionality.
• Existing methods such as Linear Discriminant Analysis (LDA) can be sensitive to these issues.
• There is a need for robust classification methods that can handle complex spectral datasets.

Purpose:

• To introduce a novel classification approach, Improved Canonical Variates Analysis-Linear Discriminant Analysis (ICVA-LDA).
• To enhance the performance of LDA by integrating it with a modified Canonical Variates Analysis (CVA) using partial least squares.
• To address the limitations of traditional methods in handling spectral data with multicollinearity and small sample sizes.

Summary:

• The ICVA-LDA model integrates an inner part estimating robust canonical variate weights via partial least squares (PLS) with an outer part building the LDA model.
• This approach ensures canonical variates are more discriminative and forces key information into fewer variates.
• The method allows interpretation in the original high-dimensional data space and demonstrates improved efficiency and conciseness.

Impact:

• The ICVA-LDA approach overcomes limitations of standard LDA, particularly with multicollinear spectral data.
• It demonstrates superior classification performance compared to PCA-LDA and standard CVA-LDA on near-infrared spectroscopy data.
• The method offers a versatile tool for classification and discrimination in various fields dealing with small samples and collinear data.