Multi-way data modelling for enhancing classification performance: Fluorescence data as a case of study.

Area of Science:

• Analytical Chemistry
• Chemometrics
• Multivariate Data Analysis

Background:

• Classification of complex systems is challenging in analytical chemistry.
• Data structure significantly impacts classification performance.
• This study investigates first- to third-order data structures using fluorescence spectroscopy.

Purpose of the Study:

• To systematically evaluate the influence of data dimensionality on classification performance.
• To compare different chemometric models including PLS-DA, N-PLS-DA, and PARAFAC-DA.
• To assess model performance under various conditions like class imbalance, noise, and sample size.

Main Methods:

• Utilized simulated and experimental fluorescence datasets.
• Employed partial least squares-discriminant analysis (PLS-DA), multi-way PLS-DA (N-PLS-DA), and PARAFAC-DA.
• Evaluated models with varying data orders (first to third).

Main Results:

• Third-order models achieved >93% accuracy, outperforming first- and second-order models.
• N-PLS-DA and PARAFAC-DA successfully discriminated olive oil samples.
• PARAFAC-DA offered superior interpretability of degradation and oxidation processes.

Conclusions:

• Higher-order data modeling, especially third-order, enhances classification reliability and interpretability.
• Third-order chemometric models are robust and generalizable for complex matrices.
• This approach offers significant potential for analytical applications with complex data.