A new and consistent parameter for measuring the quality of multivariate analytical methods: Generalized analytical...

Area of Science:

• Analytical Chemistry
• Chemometrics
• Spectroscopy

Background:

• Classical sensitivity in multivariate calibration does not fully account for noise, leading to inconsistencies in performance evaluation.
• Existing figures of merit often fail to capture the impact of diverse noise structures on prediction accuracy.
• There is a need for a robust metric that integrates noise characteristics for reliable comparison of calibration models.

Purpose of the Study:

• To introduce and define Generalized Analytical Sensitivity (γ) as a novel figure of merit for multivariate calibration.
• To demonstrate the utility of γ in comparing different calibration methodologies and resolving literature discrepancies.
• To establish the correlation between the inverse of γ and prediction errors under various noise conditions.

Main Methods:

• Development of Generalized Analytical Sensitivity (γ) incorporating noise properties.
• Application of γ to simulated and experimental first-order multivariate calibration data.
• Evaluation using Multiple Linear Regression (MLR), Principal Component Regression (PCR), and Maximum Likelihood PCR (MLPCR) models.

Main Results:

• Generalized Analytical Sensitivity (γ) effectively estimates performance by including noise characteristics.
• The inverse of γ shows a strong correlation with Root Mean Square Errors of Prediction (RMSEP) across different noise types.
• γ provides a consistent basis for comparing calibration methodologies, addressing inconsistencies found in previous studies.

Conclusions:

• Generalized Analytical Sensitivity (γ) is a superior figure of merit for multivariate calibration compared to classical sensitivity.
• γ offers a reliable method for assessing analytical methods, especially in the presence of complex noise structures.
• The proposed metric facilitates more accurate and consistent evaluation of calibration models in analytical chemistry.