A1691
Title: Poisson-gamma neural network for clustered count data
Authors: Hangbin Lee - Chungnam National University (Korea, South) [presenting]
Il Do Ha - Pukyong National University (Korea, South)
Changha Hwang - Dankook University (Korea, South)
Youngjo Lee - Seoul National University (Korea, South)
Abstract: Large-scale biomedical datasets often involve clustered count outcomes with complex correlation structures arising from high-cardinality categorical features. Standard neural network approaches typically overlook such dependencies, limiting their ability to deliver subject-specific predictions. A Poisson-gamma neural network is introduced that incorporates gamma random effects into a neural network within a hierarchical likelihood framework. The proposed model captures both nonlinear population-level patterns and subject-specific variability in a unified manner. To further improve computational efficiency and stability, an adjustment scheme for random effects and variance components is developed. Simulation studies demonstrate that the proposed model consistently outperforms existing models in terms of mean squared Pearson error and mean deviance across a range of random effect distributions. Applications to real biomedical datasets demonstrate its practical advantages for clustered count data analysis.
