Bayesian Variable Selection in Double Generalized Linear Tweedie Spatial Process Models

Double generalized linear models provide a flexible framework for modeling data by allowing the mean and dispersion to vary across observations. Common members of exponential family including Gaussian, Poisson, compound Poisson-gamma (CP-g), Gamma inverse-Gaussian are known admit such models. The lack their use can be attributed ambiguities that exist in model specification under large number covariates complications arise when display complex spatial dependence. In this work we consider hierarchical CP-g with random effect. effect is targeted at performing uncertainty quantification dependence within arising from location based indexing response. We focus on Gaussian process Simultaneously, tackle problem using Bayesian variable selection. It effected through continuous spike slab prior parameters, specifically fixed effects. novelty our contribution lies frameworks developed perform various synthetic experiments showcase accuracy frameworks. They then applied analyze automobile insurance premiums Connecticut, year 2008.