B1801
Title: Soft principal component regression using Bernstein matrix polynomials
Authors: Keith Knight - University of Toronto (Canada) [presenting]
Abstract: Ridge regression and principal component (PC) regression are useful in cases where the number of predictors exceeds the number of observations or where the predictors are highly collinear. We explore a ``compromise'' between these two methods (soft PC regression), which uses Bernstein matrix polynomials to downweight PCs with smaller variances without eliminating them. A modification of de Casteljau's algorithm is used to compute the soft PC estimates.