A1494
Title: Autonomous spatial statistical analysis with large language model agents
Authors: Tzai-Hung Wen - National Taiwan University (Taiwan) [presenting]
Abstract: Recent advances in large language models (LLMs) have accelerated task-level automation, but current AI agents remain limited in their ability to conduct rigorous spatial statistical analysis. Valid spatial inference requires more than computation; it depends on appropriate problem formulation, neighborhood specification, and diagnostic checking. An autonomous AI agent is proposed that links natural-language research questions to reproducible spatial statistical workflows under curated data settings. The key methodological contribution is a structured mapping framework that translates unstructured queries into standard operating procedures for spatial analysis. The agent integrates exploratory spatial data analysis, automated spatial weight construction, Moran's I and local cluster analysis, spatial regression modeling, and internal self-verification through geometry validation, residual diagnostics, and spatial consistency checks. The system is evaluated through two case studies: regional educational inequality and epidemiological hotspot detection. Robustness is assessed against expert-constructed benchmark workflows in terms of methodological fidelity, interpretive accuracy, and execution stability. Results show that the agent can autonomously produce end-to-end spatial statistical analyses with outputs highly consistent with benchmark expert results, suggesting a practical path toward reproducible and accessible spatial statistical modeling.
