A MODIFIED PREDICTION ERROR SUM OF SQUARES CRITERION FOR BANDWIDTH SELECTION IN LOCAL QUADRATIC REGRESSION
EDIONWE, E. AND OSEMWENKHA, S. O.
ABSTRACT
Bandwidth (smoothing parameter) is considered the most crucial parameter in the application of nonparametric regression model. For the purpose of selecting adaptive bandwidths for the Local Quadratic Regression (LQR) in the response surface settings, we present a modified Prediction Error Sum of Square (PRESS) criterion using a penalty term derived from the sum of the range of kernel weights at each of the data points. LQR is applied to a multiple response problem from the literature using the current PRESS criterion and the proposed version for optimal bandwidth selection. The proposed…