Abstract
Recently, the statisticians have developed a new approach called functional Statistics to treat the data as curves or images. In parallel, the Near-Infrared Reflectance (NIR) spectroscopy approach has been used in modern Chemistry being a fast, inexpensive and accurate procedure to characterize chemical properties for an object. In this paper, we study the forage quality by analysing the spectroscopy procedure with some modern statistical models. Our contribution leads to the prediction of chemical components of Chinese ryegrass forage by analysing its spectral data using some functional models. Precisely, the functional linear quantile regression (FLQR), the functional nonparametric quantile regression (FNQR), the functional local linear quantile regression (FLLQR) and the functional local linear model regression (FLLMR) are implemented to predict the quantities of acid detergent fiber (ADF), neutral detergent fiber (NDF), and crude protein (CP) contents. The choice of these functional models is motivated by the fact that they can construct a predictive region with a given confidence level. We show that the considered models improve the prediction results significantly as compared to conventional models such as the classical partial least squares regression (PLSR) and the principal component regression (PCR). Moreover, we also show that the proposed models are more robust than their competitive models like PLSR and PCR in the sense that their efficiency is not much affected by non homogeneity of the data.