Local polynomial additive regression through PLS and splines: PLSS

We present a recently devised extension of the linear Partial Least Squares (PLS) model to the nonlinear additive through the transformation of predictors by polynomial spline functions that we call Partial Least Squares Splines (PLSS). Suitable tuning parameters of PLSS allow the user to experiment with a wide range of PLS regression tools, from the classical linear and polynomial models towards more flexible local polynomial additive modeling. Due to B-spline basis functions, PLSS models are not very sensitive to extreme values of the predictors in contrast to most component-based regressions. This paper aims at presenting the method like a userʹs guide and a real example of sensory analysis illustrates the performance of PLSS in the presence of outliers and nonlinear relationships.