Comparison of Ensemble Strategies in Online NIR for Monitoring the Extraction Process of Pericarpium Citri Reticulatae Based on Different Variable Selections

Different ensemble strategies were compared in online near-infrared models for monitoring active pharmaceutical ingredients of Traditional Chinese Medicine. Bagging partial least square regression and boosting partial least square regression were adopted to near-infrared models, to determine hesperidin and nobiletin content during the extraction process of Pericarpium Citri Reticulatae in a pilot scale system. Different pretreatment methods were investigated, including Savitzky-Golay smoothing, derivatives, multiplicative scatter correction, standard normal variate, normalize, and combinations of them. Two different variable selection methods, including synergy interval partial least squares and backward interval partial least squares algorithms, were performed. Based on the result of the synergy interval partial least squares algorithm, bagging partial least square regression and boosting partial least square regression were adopted into the quantitative analysis. The results demonstrated that the established approach could be applied for rapid determination and real-time monitoring of hesperidin and nobiletin in Pericarpium Citri Reticulatae (Citrus reticulata) during the extraction process. Comparing the results, the boosting partial least square regression provided a slightly better accuracy than the bagging partial least square regression. Finally, this paper provides a promising ensemble strategy on online near-infrared models in Chinese medicine.