Global Clustering Quality Coefficient Assessing the Efficiency of PCA Class Assignment

Area of Science:

• Analytical Chemistry
• Chemometrics
• Data Science

Background:

• Principal Component Analysis (PCA) is crucial for predictive modeling, but its efficiency hinges on data clustering quality.
• Score plot cluster position and dispersion significantly impact class assignment sensitivity and selectivity.

Purpose of the Study:

• To introduce objective, quantitative indicators for assessing data clustering quality in PCA.
• To propose a Global Clustering Quality Coefficient (GCQC) for evaluating PCA model predictive power.

Main Methods:

• Development of analytical geometry-inspired indicators for cluster assessment.
• Application of these indicators to evaluate PCA-based screening of amphetamines using GC-FTIR spectra.
• Validation using estimated density distributions and Quadratic Discriminant Analysis (QDA).

Main Results:

• The proposed indicators provide an objective measure of clustering quality.
• The GCQC effectively ranks preprocessing functions for PCA model efficiency.
• GCQC rankings correlate with density distributions and QDA validation.

Conclusions:

• The developed indicators and GCQC offer a robust framework for evaluating PCA model performance.
• This quantitative approach enhances the reliability of PCA in chemical analysis and predictive modeling.
• Objective assessment of clustering quality is vital for efficient PCA-based classification systems.