A distribution-free criterion for variable selection in semiparametric spatial zero-inflated models
Аннотация
Analyzing spatial zero-inflated count data often requires flexible modeling frameworks to address the challenges of excess zeros and spatial dependence. Semiparametric spatial zero-inflated models provide a robust alternative by avoiding strong distributional assumptions on the response variable. However, model selection within this semiparametric framework remains a critical and unresolved issue due to the lack of appropriate criteria that are both distribution-free and computationally feasible. In this work, we propose a novel variable selection approach for semiparametric spatial zero-inflated models based on a distribution-free criterion. The proposed criterion enables efficient identification of covariates associated with the frequency component without requiring a fully specified likelihood, thereby substantially reducing computational burden. Extensive simulation studies demonstrate that the proposed method achieves high model selection accuracy and strong robustness across varying spatial correlation levels and zero-inflation intensities, while maintaining substantial computational efficiency. An application to Taiwan’s 2016 daily extreme rainfall data further illustrates the practical utility and scalability of the proposed approach.