PREDICTIVE INFORMATION CRITERIA FOR BAYESIAN NONLINEAR REGRESSION MODELS
Daeju Kim, Shuichi Kawano, Sadanori Konishi · Bulletin of informatics and cybernetics · 2012
Bayesian nonlinear regression modeling based on basis expansions provides efficient methods for analyzing data with complicated structure. A crucial issue in the model building process is the choice of adjusted parameters including hyper-parameters for prior distribution and the number of basis functions. Choosing these parameters can be viewed as a model selection and evaluation problem. We present an information criterion for evaluating Bayesian nonlinear regression models. Our proposed modeling procedure enables us to select the appropriate values of hyper-parameters and the number of basis functions. We use a real data analysis and simulation studies to validate the performance of the proposed nonlinear regression modeling. Simulation studies show that our proposed modeling strategy performs well in various situations.