Averaging and Stacking Partial Least Squares Regression Models to Predict the Chemical Compositions and the Nutritive Values of Forages from Spectral Near Infrared Data

Partial least square regression (PLSR) is a reference statistical model in chemometrics. In agronomy, it used to predict components (response variables y) of chemical composition vegetal materials from spectral near infrared (NIR) data X collected spectrometers. PLSR reduces the dimension by defining vectors that are then as latent (LVs) multiple linear model. One difficulty determine relevant dimensionality (number LVs) for given data. This step can be very time consuming when many datasets have processed and/or frequently updated. The paper focuses on an alternative, bypassing determination and allowing automatizing predictions. strategy uses ensemble learning methods, such averaging or stacking predictions set models with different dimensionalities. presents various methods compares their performances usual six real types forages. main finding study was overall superiority compared PLSR. We therefore believe recommended analyze NIR