A novel multivariate regression approach based on kernel partial least squares with orthogonal signal correction

This paper introduces a novel multivariate regression approach based on kernel partial least squares (KPLS) with orthogonal signal correction (OSC). OSC has been proposed as a data preprocessing method that removes from X information not correlated to Y. KPLS is a promising regression method for tackling nonlinear systems because it can efficiently compute regression coefficients in high-dimensional feature spaces by means of nonlinear kernel functions. Unlike other nonlinear partial least squares (PLS) techniques KPLS does not entail any nonlinear optimization procedures and has a complexity similar to that of linear PLS. In this paper, the prediction performance of the proposed approach (OSC-KPLS) is compared to those of PLS, OSC-PLS and KPLS using three examples. OSC-KPLS effectively simplifies both the structure and interpretation of the resulting regression model and shows superior prediction performance compared to linear PLS.