Breaking the Bottleneck in Spatial Model Selection: Developing an Efficient Distribution-Free Model Selection Method
- 3 days ago
- 1 min read
Professor Chun-Shu Chen and his student from the Graduate Institute of Statistics at National Central University have developed a novel variable selection method for spatial zero-inflated data, overcoming the limitations of conventional approaches that rely on specific distributional assumptions and computationally intensive procedures. Their proposed distribution-free model selection criterion integrates semiparametric spatial zero-inflated modeling with low-rank approximation, achieving strong variable selection accuracy while substantially improving computational efficiency. The method was successfully applied to extreme rainfall data in Taiwan to identify key environmental factors associated with the occurrence and frequency of extreme rainfall events. This research demonstrates the potential of advanced statistical methodology for large-scale spatial data and climate risk analysis. The study was published in Environmental and Ecological Statistics, a Q1 journal in the field of statistics (https://doi.org/10.1007/s10651-026-00756-z) in this August.
