Calibration Model Updating to Novel Sample and Measurement Conditions without Reference Values.

Area of Science:

• Analytical Chemistry
• Chemometrics
• Spectroscopy

Background:

• Updating calibration models to new conditions (secondary domain) is crucial in analytical chemistry to avoid complete recalibration.
• Existing methods require labeled secondary samples, which are time-consuming and expensive to obtain.
• A need exists for model updating methods that do not require labeled secondary samples, enabling on-demand updates.

Purpose of the Study:

• To compare analytical model updating methods with and without labeled secondary samples.
• To develop and evaluate a hybrid model updating approach.
• To assess a framework for automatic model selection in unlabeled model updating.

Main Methods:

• Comparison of model updating techniques using labeled and unlabeled secondary samples.
• Development and evaluation of a hybrid model updating strategy.
• Application of a model selection framework based on model diversity and prediction similarity (MDPS) for unlabeled samples.

Main Results:

• The MDPS framework effectively automates model selection for updating methods without secondary analyte reference values.
• Updated models formed and selected on demand using MDPS eliminate the need for complex cross-validation.
• MDPS selected reliable updated models that matched or surpassed prediction errors from full recalibrations using labeled data across four near-infrared datasets.

Conclusions:

• The MDPS framework provides an efficient and reliable solution for selecting updated analytical models without requiring labeled secondary samples.
• This approach significantly reduces the cost and time associated with recalibration by enabling on-demand model adaptation.
• The method demonstrates strong performance, rivaling traditional recalibration techniques and offering a practical alternative for analytical laboratories.