ScholarWorks@성균관대학교

초록

Count data with quasi-sparsity, which consist of a few signals and an overabundance of small counts including zeros, are prevalent across diverse domains such as crime statistics, epidemiology, and genetics. However, existing methods for handling zero-inflated counts are not sufficient to simultaneously detect the signal and the noise. This paper introduces a Bayesian multivariate spatio-temporal model for quasi-sparse count data, incorporating a shrinkage effect into a P-MSTM framework proposed by Bradley et al. (2018). The shrinkage effect with global-local shrinkage priors globally shrinks the noise towards zero and locally allows the large counts. The simulation studies demonstrate the superior performance of the proposed method. For application, we apply the model to the crime data in Los Angeles and predict day-to-day occurrence.

키워드