Gaussian process regression for prediction and confidence analysis of fruit traits by near-infrared spectroscopy

Abstract Detection of fruit traits by using near-infrared (NIR) spectroscopy may encounter out-of-distribution samples that exceed the generalization ability a constructed calibration model. Therefore, confidence analysis for given prediction is required, but this cannot be done common models NIR spectroscopy. To address issue, paper studied Gaussian process regression (GPR) detection The mean and variance GPR were used as predicted value confidence, respectively. show this, real data set related to dry matter content measurements in mango was used. Compared partial least squares (PLSR), showed approximately 14% lower root squared error (RMSE) in-distribution test set. with no analysis, remove abnormal made PLSR 58% 10% RMSE on mixed distribution set, respectively (when type 1 rate 0.1). traditional one-class classification methods, can effectively eliminate poorly samples.