Machine learning-based mapping of indoor radon potential using geogenic factors

This study presents a comprehensive, data-driven analysis of mapping indoor radon potential areas in the Apulia region of southeastern Italy, relying exclusively on geo- genic factors. Key geogenic radon factors intrinsic to the Earth’s geological makeup are examined, including the presence of uranium in bedrocks and soil, geological inhomogeneities, such as faults and karstic formations, and soil properties affect- ing gas permeability. The methodology utilizes Frequency Ratio (FR) analysis to empirically derive the relevance of 14 geogenic subfactors which are then used as standardized inputs for Machine Learning Algorithms including Support Vector Machine (SVM), Multilayer Perceptron Artificial Neural Network (MLP-ANN), and Random Forest. Each model is trained and validated using a balanced dataset derived from indoor radon measurements and geospatial geogenic predictors. SHAP (SHapley Additive exPlanations) analysis is used to enhance interpretability by showing how geogenic factors influenced model predictions. Model performance is evaluated using confusion matrices, ROC (Receiver Operator Characteristic) curves, and AUC (Area Under the Curve) metrics. Notably, SVM emerges as the top-performing model among them. Key findings highlight the significant influence of geological inhomogeneities, uranium content in soil and bedrock, as well as soil permeability on indoor radon concentration. The proposed workflow combining FR and SHAP for interpretability demonstrates an effective and reproducible strategy for regional indoor radon potential mapping. The research contributes to advancing the understanding of indoor radon exposure dynamics, providing valuable insights for policymakers, urban planners, and public health authorities to mitigate radon- related health risks effectively by identifying high indoor radon potential areas based on geogenic factors and elucidating underlying mechanisms; this study prepares the groundwork for targeted intervention strategies and informed decision-making to safeguard public health and enhance community resilience against radon exposure hazards in the Apulia region. Based on data from extensive geospatial analysis, this framework offers a reliable approach for regional radon potential mapping in other data-scarce regions as well.