Abstract:
|
In this article, we consider the the Bayesian multi-index additive model. The index is modeled by the polar coordinates and the function of each additive component is approximated by the Bayesian B-splines. We developed the Markov chain Monte Carlo algorithm to sample the parameters from the corresponding posterior distribution. Bayesian information criterion is applied to choose the number of indexes and the number of knots in the B-splines. It has been shown that through both simulation and real data analysis that the proposed method works better than existing methods, such as MAVE, random forest, and others.
|