Pattern classification adopting multivariate polynomials

The use of a full multivariate polynomial model for predictor learning was deemed a daunting task due to its explosive number of expansion terms for high dimensional inputs and high order models. This paper investigates into the viability of using full multivariate polynomials for predictor learning. Particularly, we investigate into the frequently encountered under-determined system with an estimation formulation based on a ridge regression beyond the commonly known primal and dual forms. Extensive experiments are performed to observe the predictor learning properties on polynomial models beyond the frequently adopted second order.