SUBSET SELECTION IN NONLINEAR POISSON REGRESSION USING SUPPORT VECTOR REGRESSION : A SIMULATION STUDY

Subset selection in linear Poisson regression model is a well studied problem in the Literature. It is established that distribution function criterion (DFC) based on the MLE of regression parameters is a consistent subset selection criterion in generalized linear models (GLM). Nonlinear Poisson regression model is a novel generalization of linear Poisson regression model. In this paper, we propose the use of distribution function criterion (DFC) for subset selection in nonlinear Poisson regression model. To compute DFC, we propose the use of support vector regression (SVR). An extensive simulation study to compare the performance of DFC based on MLE (GLM fit) and SVR for subset selection in linear and nonlinear Poisson regression is carried out. The results reveal that DFC based on SVR is almost twice as efficient as DFC based on MLE (GLM fit) in nonlinear Poisson regression model when the exact functional form of the relationship is unknown.