A new approach for the identification of important variables

Combined the sequential projection pursuit and Partial Least Squares (PLS) analysis, a novel approach was firstly proposed for the identification of the most important variables. The computational experiments indicated that the approach proposed could obtain more satisfactory analytical results than that of PLS analysis and stepwise regression. Moreover, this work indicated that the prediction capability of the model was not always suitable to be the criterion of the identification of the most important variables.