A novel physics-informed AI framework for the assessment and prediction of indoor radon concentration and risk classification.

Indoor radon gas is a leading environmental cause of lung cancer, yet accurate risk assessment remains challenging due to the practical difficulties of direct measurement. This study introduces a novel Physics-Informed Neural Network (PINN) framework that integrates physical laws of radon transport with machine learning to predict indoor radon concentrations (Qt). Our Geologically-Informed Radon Assessment (GIRA) model incorporates radon contributions from geological foundations (Qg), faults (Qf), and building materials (Qb), while accounting for building porosity. When validated against a dataset of 957 structures in Western Türkiye, the PINN model significantly outperformed conventional machine learning approaches, achieving a Mean Absolute Error of 52 Bq/m3 and R2 of 0.96. The framework successfully identified 15.3% of structures as high-risk (>300 Bq/m3), demonstrating its capability for automated radon risk classification. This physics-informed approach provides a robust, interpretable, and cost-effective tool for proactive public health planning and targeted radon mitigation strategies, establishing a new paradigm in environmental hazard assessment.