This study focuses on one-step ahead radon time-series forecasting using Long-Short Term Memory (LSTM). The predictor is based on a n-size sliding window on a multivariate time series containing soil radon data from the Phlegraean Fields area and various geophysical parameters. After evaluating a linear model as a base, using metrics such as mean absolute error and mean square error, we show that an LSTM model, which also exploits imputed information, exhibiting superior capabilities in capturing trends and peaks. This results shows better performance, including an improved Pearson’s correlation coefficient, compared previous studies.
LSTM-Based Models for Radon Forecast / Di Giovanni, M., Palmieri, F.A.N., Di Gennaro, G., Buonanno, A., Ambrosino, F., Pugliese, M., Verde, G.L., Sabbarese, C.. - 459:(2026), pp. 235-245. (32nd International Workshop on Neural Networks, WIRN 2024 ita 2024) [10.1007/978-981-95-4072-3_20].
LSTM-Based Models for Radon Forecast
Ambrosino, Fabrizio;Pugliese, Mariagabriella;Verde, Giuseppe La;Sabbarese, Carlo
2026
Abstract
This study focuses on one-step ahead radon time-series forecasting using Long-Short Term Memory (LSTM). The predictor is based on a n-size sliding window on a multivariate time series containing soil radon data from the Phlegraean Fields area and various geophysical parameters. After evaluating a linear model as a base, using metrics such as mean absolute error and mean square error, we show that an LSTM model, which also exploits imputed information, exhibiting superior capabilities in capturing trends and peaks. This results shows better performance, including an improved Pearson’s correlation coefficient, compared previous studies.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
