This paper reports the analysis of soil 222Rn data recorded over 7-years in the volcanic caldera of Campi Flegrei (Naples-Italy). The relationship between Radon activity concentration and several geophysical, geochemical and meteorological parameters, influencing the gas emissions, is estimated by the Artificial Neural Network (ANN) method. The analysis goals are: the estimation (replication) of the Radon time series from influencing parameters, the forecasting of an unknown part of it, and the search for anomalies. Results prove: (i) the effectiveness of the ANN method; (ii) Radon follow the periods of agitation of the caldera, demonstrated by the comparison with previous works using different methods. •Radon time series at Campi Flegrei volcanic caldera is analysed.•Radon is estimated by Artificial Neural Network trained with influencing parameters.•The forecasting of an unknown part and the anomalies detection are performed.•The results prove high performance in estimation and forecasting of Radon signal.•Present anomalies coincide with anomalies obtained in papers using a different method.