Coffee roasting degrees prediction in terms of weight loss with selected wavebands based on near-infrared spectroscopy

Waveband selection was investigated to improve a coffee roasting degrees model in terms of weight loss. Near infrared diffuse reflectance spectra were obtained from ground roasted coffee samples. Characteristic wavebands were determined by regression coefficients of optimal principle components (PCs) of a partial least square (PLS) regression model. After that, a multiple linear regression (MLR) model was developed based on the selected wavebands. For comparison, PLS models based on different preprocessing spectra were also established. The model results of PLS and MLR method were compared. The promising result of the MLR model (Rcv2 of 0.983 and RMSECV of 0.612) indicated that waveband selection could improve regression model performance and reduce complexity, compared to the PLS regression model.