Workflow for the Supervised Learning of Chemical Data: Efficient Data Reduction-Multivariate Curve Resolution...

Area of Science:

• Chemometrics
• Data Science
• Machine Learning

Background:

• High-dimensional data classification presents challenges in variable selection and training set optimization.
• Existing methods often struggle with small sample sizes and class imbalances.

Purpose of the Study:

• To introduce and evaluate the novel efficient data reduction-multivariate curve resolution (EDR-MCR) method for supervised classification of high-dimensional data.
• To demonstrate the efficacy of coupling EDR and MCR for data splitting, variable selection, and classification.

Main Methods:

• The EDR-MCR method combines efficient data reduction (EDR) using principal component analysis (PCA) and convex geometry with multivariate curve resolution (MCR) for classification.
• Data dimensionality is reduced, and training sets are selected before classification using an MCR model with imposed numerical constraints.

Main Results:

• EDR demonstrated comparable performance to other data splitting methods, even with smaller training sets.
• The proposed MCR approach showed advantages in speed, parameter tuning, and flexibility for data with low sample numbers and class imbalances.
• EDR-MCR improved accuracy by incorporating system information through numerical constraints and resolved pure component signal weights.

Conclusions:

• EDR-MCR offers an effective and efficient strategy for supervised classification of high-dimensional data.
• The method provides advantages over traditional techniques, particularly for complex datasets with limited samples or imbalanced classes.
• EDR-MCR facilitates the extraction of interpretable pure component information from complex data mixtures.