A COMPARATIVE STUDY OF SEMIPARAMETRIC REGRESSION MODELS: A NOVEL BLEND VIA LOCALLY ADAPTIVE BANDWITHS SELECTOR FOR RESPONSE SURFACE METHODOLOGY (RSM

Abstract

Response Surface Methodology (RSM) is a sequential statistical approach employed by engineers and industrial
statistician for empirical model building where the processes and products are optimized. In RSM, the parametric
regression models are frequently used but lack credibility due to model misspecification and as such affects the
process mean and variance and ultimately, the estimated response is misscalulated. Although, the nonparametric
regression models are flexible, but lack recognition in RSM due to the idiosyncracies of RSM data; such as
dimensionality problem, sparseness of RSM data and small sample size. In the literature, semiparametric
regression models are considered the most suitable methods in RSM because it combines attributes of the
parametric and nonparametric regression models in a fashionable manner. In this paper, we give a comparative
analysis of the OLS and three semiparametric regression models that utilizes two existing locally adaptive
bandwidths from the literature were applied to obtain a novel blend of the semiparametric regression models used
to smooth the two data for the application problems and the results tend to improve the goodness of fit statistics,
with minimum residual plots for the responses and optimization of processes and products.