Comparative analysis of preprocessing methods and machine learning models for geographical origin identification in an imbalanced Panax notoginseng dataset with near-infrared spectroscopy

Imbalanced sample distributions pose challenges in accurately identifying the geographical origin of <i>Panax notoginseng</i> (<i>P. Notoginseng</i>), impacting machine learning model performance. This study investigates the efficacy of various data preprocessing methods and machine learning models on geographical origin classification accuracy using near-infrared spectroscopy (NIRS) in imbalanced datasets. Results demonstrate that specific combinations of preprocessing methods and models can significantly improve classification accuracy, bridging the gap between imbalanced and balanced datasets. These findings highlight the importance of tailored approaches to mitigate model and evaluation bias, imbalanced sample weights, and decision boundary shifts in <i>P. Notoginseng</i> datasets. Moreover, the insights gleaned from this study have implications for similar analyses in other traditional Chinese medicinal materials.