Combining partial least squares regression and least squares support vector machine for data mining

摘要

PLS can effectively eliminate the multicolinearity among explanatory variables and LSSVM can reflect the nonlinear relations between dependent variable and explanatory variables. PLS and LSSVM are combined together. In PLS-LSSVM model, PLS is used to extract the independent components, then the extracted components is input to the LSSVM with radial basis kernel function for predicting. The LSSVM parameters are determined by cross validation based on grid search. The experiment results of PLS-LSSVM are compared with partial least squares regress, which show that PLS-LSSVM model can be trained quickly and has good generalization.