Cluster resolution: a metric for automated, objective and optimized feature selection in chemometric modeling

Area of Science:

• Chemometrics
• Data Science
• Machine Learning

Background:

• Chemometric models require robust feature selection for optimal performance.
• Automated feature selection methods are crucial for handling high-dimensional data.
• Existing methods may not fully account for cluster geometry and orientation.

Purpose of the Study:

• Introduce a novel metric, cluster resolution, for evaluating data cluster separation.
• Develop a fully automated feature selection process for chemometric model construction.
• Demonstrate the utility of cluster resolution in classifying gasoline samples by octane rating.

Main Methods:

• The cluster resolution metric quantifies cluster separation considering shape and orientation via confidence ellipses.
• Principal Component Analysis (PCA) was used for sample classification.
• Automated variable ranking (ANOVA) and forward selection were employed for feature selection.

Main Results:

• The cluster resolution metric effectively evaluates cluster separation and overlap.
• Automated feature selection using cluster resolution enabled objective model construction.
• PCA successfully classified gasoline samples based on octane rating using selected features.

Conclusions:

• Cluster resolution is a valuable metric for automated feature selection in chemometrics.
• This approach facilitates the development of objective and fully automated chemometric models.
• The method shows broad applicability across various data analysis and modeling tasks.