Developing of calibration model for acetic acid, flavonoid, and capsaicin content from fresh red chilies using combination Vis-NIR spectral descriptors, machine learning and their stacking ensemble learning

Abstract

Rapid and non-destructive measurements to determine the amount of certain compounds in products such as fresh red chilies are still challenging today. Therefore, the objective of this study was to develop a fast and non-destructive calibration model of fresh red chilies for the content of acetic acid, flavonoid, and capsaicin using a combination of visible near-infrared spectroscopy (Vis-NIR) descriptors in the range 380–1065 nm and machine learning regressors including random forest (RF), Ridge, decision tree (DT) and their stacking ensemble learning (SEL). The best calibration model to predict the content of acetic acid and flavonoid compounds is to use SEL, and capsaicin is to use DT. Thus, the proposed model can effectively predict the compound content in fresh red chilies and may have important applications in helping researchers and practitioners assess its quality non-subjectively and non-invasively.

Keywords

• Agriculture product
• Chemometrics
• Non-destructive measurements
• Machine learning