A training set selection strategy for a universal near-infrared quantitative model.

Area of Science:

• Analytical Chemistry
• Chemometrics
• Spectroscopy

Background:

• Developing accurate quantitative models for complex samples is challenging.
• Near-infrared (NIR) spectroscopy offers a rapid, non-destructive analytical technique.
• Optimizing training set selection is crucial for robust NIR model performance.

Purpose of the Study:

• To propose an empirical method for determining the optimal number of clusters for training set construction.
• To develop a strategy for building universal near-infrared quantitative models using the Naes method.
• To ensure efficient and accurate model development for complex sample analysis.

Main Methods:

• Hierarchical clustering of sample spectra using Ward's algorithm and Euclidean distance.
• Determining the optimal number of clusters based on heterogeneity values.
• Randomly selecting one sample per cluster to form the training set.
• Applying the Naes method for quantitative model development.

Main Results:

• The proposed strategy was applied to 98 batches of rifampicin capsules, achieving root mean square errors of cross-validation and prediction of 2.54% and 2.31%, respectively.
• The model demonstrated satisfactory outlier diagnostics, accuracy, precision, and robustness.
• Revalidation on cefradine capsules, roxithromycin tablets, and erythromycin ethylsuccinate tablets yielded positive results.

Conclusions:

• The developed training set sample selection strategy enables rapid and accurate construction of quantitative models.
• This approach is effective for near-infrared spectroscopy applications across various pharmaceutical products.
• The method provides a reliable solution for optimizing complex sample analysis in quantitative modeling.