Identifying Healthcare Access and Enhancing Geospatial Analysis with Generative AI

Abstract

This thesis brings together geospatial modeling and generative artificial intelligence to address healthcare accessibility and automation of spatial analysis. The first study examines disparities in dental care access across six regions of Virginia by comparing driving and public transit modes for both all dental clinics and those accepting Medicaid. Using a modified two-step floating catchment area (2SFCA) method, the research quantifies access based on travel time, supply-demand ratios, and vehicle ownership. Results show that public transit accessibility is significantly lower and more unequal than driving access, particularly for Medicaid recipients, with variation across regions. Spatial error models further reveal demographic factors, such as poverty, race, and vehicle access, influence accessibility patterns. The second study fine-tunes OpenAI's GPT-4o-mini model to convert natural language queries into executable Python code for geospatial analysis. Trained on over 600 geospatial prompt-completion pairs using Virginia health data, the model achieves an 89.7% accuracy rate, improving significantly over the baseline. It integrates spatial reasoning, fuzzy geographic matching, and modular function calls to reduce execution errors and enhance usability. Together, these studies demonstrate how AI and geospatial science can jointly address inequities in healthcare access while making spatial tools more accessible to policymakers, researchers, and the general public.