A0205
Title: Bayesian region selection and prediction in Poisson regression with spatially dependent global-local shrinkage prior
Authors: Zihan Zhu - Case Western Reserve University (United States) [presenting]
Xueying Tang - University of Arizona (United States)
Shuang Zhou - Arizona State University (United States)
Abstract: A spatially dependent global-local shrinkage prior is proposed for Poisson regression, specifically aimed at prediction and region selection with spatially dependent covariates. This approach is inspired by the challenge of predicting the number of hurricanes and identifying regions with significant contributions based on spatially dependent data. The proposed prior combines the conditional autoregressive (CAR) prior, which introduces spatial dependence in the coefficients of spatially dependent covariates, with the super heavy-tailed (SH) prior, which ensures appropriate global-local shrinkage effects for selection. Metropolis-within-Gibbs sampler is developed for computation. Extensive simulation studies demonstrate that the method excels when signals are weak and adjacent, and the spatial dependence in covariates is strong. Applied to North Atlantic hurricane prediction, the method outperforms traditional regression-based approaches and rivals the benchmark "oracle" model.
