Paper Type
Complete
Abstract
Designing spatial risk decision-support systems involves more than improving predictive accuracy. In many decision-support applications, outcomes such as health burden or environmental risk are multidimensional, reflecting multiple related indicators rather than a single measure. Without explicit design controls, models may appear accurate while producing unstable or geographically distorted guidance. This study develops a five-layer explainable GeoAI architecture for spatial risk modeling. The framework specifies sequential controls: spatial data integration, composite outcome construction, structured feature governance, multi-model estimation, and decision validation using cross-model stability checks and spatial diagnostics. The architecture is demonstrated using neighborhood-level mental health burden across 8,850 California census tracts. Results show strong explanatory and predictive performance (R² = 0.85; ROC-AUC > 0.98) and consistent determinant structure across linear and ensemble models. Spatial diagnostics reveal residual clustering and regional variation in coefficient strength, indicating that global assumptions can mask local heterogeneity. The contribution is architectural, embedding governance, interpretability, and spatial validation directly into scalable decision-support systems that are transferable to other place-based risk contexts.
Paper Number
1108
Recommended Citation
Satpathy, Asish and Lingaraju, Deeksha, "Designing Explainable GeoAI Systems for Spatial Risk Decision Support" (2026). AMCIS 2026 Proceedings. 1.
https://aisel.aisnet.org/amcis2026/sig_dsa/sig_dsa/1
Designing Explainable GeoAI Systems for Spatial Risk Decision Support
Designing spatial risk decision-support systems involves more than improving predictive accuracy. In many decision-support applications, outcomes such as health burden or environmental risk are multidimensional, reflecting multiple related indicators rather than a single measure. Without explicit design controls, models may appear accurate while producing unstable or geographically distorted guidance. This study develops a five-layer explainable GeoAI architecture for spatial risk modeling. The framework specifies sequential controls: spatial data integration, composite outcome construction, structured feature governance, multi-model estimation, and decision validation using cross-model stability checks and spatial diagnostics. The architecture is demonstrated using neighborhood-level mental health burden across 8,850 California census tracts. Results show strong explanatory and predictive performance (R² = 0.85; ROC-AUC > 0.98) and consistent determinant structure across linear and ensemble models. Spatial diagnostics reveal residual clustering and regional variation in coefficient strength, indicating that global assumptions can mask local heterogeneity. The contribution is architectural, embedding governance, interpretability, and spatial validation directly into scalable decision-support systems that are transferable to other place-based risk contexts.
When commenting on articles, please be friendly, welcoming, respectful and abide by the AIS eLibrary Discussion Thread Code of Conduct posted here.
Comments
SIG DSA