A machine learning framework for radon forecasting with uncertainty quantification in the Campi Flegrei volcanic area

This study presents a machine learning framework for short-term forecasting of radon activity concentration in the Campi Flegrei volcanic area. The dataset integrates fourteen years of radon, meteorological, and seismic observations through a dedicated preprocessing pipeline that manages missing values, uneven sampling, and aggregation into 3-hour intervals. Two model families are compared: tree-based (XGBoost) and sequence-based (LSTM). Conformal prediction is employed to generate statistically valid uncertainty intervals, complementing deterministic forecasts. Results show that XGBoost achieves the highest point-prediction accuracy, while LSTM demonstrates greater robustness in recursive multi-step forecasts. The proposed framework enhances the reliability of environmental time-series forecasting and provides a reproducible tool for exploring the coupling between radon emissions, meteorological variability, and seismic activity in volcanic regions.