FairGNN-Climate: Fairness-Constrained Graph Neural Networks for Urban Neural Resilience Under Climate Stress

Rapid urbanization and climate change are increasing environmental stressors such as heat waves, air pollution, and noise in densely populated cities. These stressors disproportionately affect vulnerable communities and are closely linked to neurological and mental health challenges, including anxiety, cognitive strain, sleep disruption, and reduced resilience. However, most climate and health prediction models focus mainly on improving accuracy and often overlook fairness, which can unintentionally reinforce existing social inequalities embedded within urban systems and public infrastructure.

This study presents FairGNN Climate, a graph based framework designed to predict urban neural resilience under climate stress while incorporating fairness considerations directly into the learning process. The system represents urban populations as interconnected graphs that combine environmental exposure, socioeconomic conditions, health indicators, and demographic information. By modeling relationships between neighborhoods and shared climate exposure, the framework captures patterns of correlated vulnerability that traditional independent sample models often miss. Fairness measures, including demographic parity and equalized odds, are integrated during training to reduce disparities across different population groups.

The framework is developed as a complete decision support platform that enables resilience prediction, fairness evaluation, geospatial vulnerability analysis, and counterfactual policy simulation. Applied to real urban data from Indian cities, the results demonstrate that the model produces interpretable resilience estimates and highlights structural disparities that require continued monitoring, responsible governance, and sustained policy level attention.