Selection of the best calibration sample subset for multivariate regression.

Area of Science:

• Analytical Chemistry
• Chemometrics

Background:

• Principal Component Regression (PCR) analysis often requires a large number of calibration samples, increasing costs and time.
• Optimizing the selection of calibration samples is crucial for efficient analytical method development.

Purpose of the Study:

• To develop and validate a methodology for selecting the minimum number of calibration samples in PCR analysis.
• To reduce the costs associated with chemical analysis and calibration without compromising accuracy.

Main Methods:

• A novel method utilizing only instrumental responses of a large sample set to select an optimal subset for calibration.
• The subset selection aims to minimize the variance of regression coefficients.
• Application of the methodology to UV-visible spectroscopy for Ca(2+) determination and near-IR spectroscopy for moisture determination in corn.

Main Results:

• Regression models developed with the reduced subset of calibration samples yielded accurate results in both spectroscopic applications.
• The precision, measured by root-mean-squared error of cross-validation (RMSECV), was comparable to models using the complete calibration set.
• The number of analyzed samples in the calibration set was reduced by up to 50%.

Conclusions:

• The proposed methodology effectively reduces the number of calibration samples needed for PCR analysis.
• This approach offers significant cost savings in chemical analysis and calibration.
• The method ensures accurate and precise analytical results, demonstrating its practical utility in spectroscopy.