Optimizing calibration sample selection in infrared spectroscopy
Résumé
• Models trained on samples selected by the KS and WSP algorithms performed better than those trained on randomly selected samples.
• Both WSP and KS generally benefit from dimensionality reduction and encoding or at least maintain the predictive performance of the models.
• For higher-complexity data, differences in the predictive performance of models were more pronounced between selection algorithms.
• Preprocessing affects the predictive performance of models trained on samples selected by WSP and KS differently.