Predictive model selection criteria for relevance vector regression models
和己 松田 · Josai University Repository of Academia (Josai University) · 2015
We focus on a selection of kernel parameters in the framework of the relevance vector machine (RVM) for regression, called the relevance vector regression (RVR). The RVR can achieve a sparse model and utilize a kernel function similar to the support vector regression (SVR). A crucial issue in the model building process of the RVR is the selection of the optimal values for kernel parameters. ln this paper, we derive a model selection criterion for evaluating the Bayesian predictive distribution of the RVR model from information-theoretic viewpoint. Monte Carlo experiments and real data analysis have been presented to demonstrate that the proposed modeling procedure performs well.