GeoXCP: Uncertainty Quantification in Spatial Explanations from Explainable AI

This repository contains the code and data used in our submitted work “GeoXCP: Uncertainty Quantification in Spatial Explanations from Explainable AI.”In this study, we propose GeoXCP, a model-agnostic framework for quantifying uncertainty in geospatial explanations derived from explainable AI (XAI) methods.We validate the effectiveness of GeoXCP through a case study on housing price prediction. The provided files are organized as follows:📁 1-modelContains the core implementation of GeoXCP, including the following modules:•GeoConformal•GeoConformalizedExplainer📁 2-data_Housing_priceIncludes the housing price dataset for Seattle:•seattle_sample_3k.csv📁 3-codes_Housing_priceContains code used for applying GeoXCP to the housing price prediction task:•SeattleHomeSaleUncertaintyGeoX.ipynb: End-to-end implementation for uncertainty quantification using GeoXCP. All results from the main experiment can be reproduced here.•seattle_house_price_coverage_bootstrap.ipynb: Demonstrates how GeoXCP achieves reliable coverage through bootstrap-based validation.📁 4-codes_simulation_studyContains code used for applying GeoXCP to the simulated dataset:•simulation_study_linear_SHAP.ipynb: Evaluate GeoXCP with linear function and SHAP.•simulation_study_nonlinear_SHAP.ipynb: Evaluate GeoXCP with nonlinear function and SHAP.•simulation_study_linear_LIME.ipynb: Evaluate GeoXCP with linear function and LIME.