GNGTS 2023 - Atti del 41° Convegno Nazionale

Session 2.2 GNGTS 2023 Liquefaction probability prediction with machine learning models based on penetrometric investigations C. Varone 1 , F. Mori 1 , A. Mendicelli 1 , L. Martelli 2 , G. Naso 3 , M. Moscatelli 1 1 CNR Italian National Research Council, Institute of Environmental Geology and Geoengineering (IGAG), RM1, 00015 Montelibretti, Italy 2 Geological, Seismic and Soil Survey, Emilia-Romagna Region, Bologna, 40127, Italy 3 Presidenza del Consiglio dei Ministri, Dipartimento della Protezione Civile (DPC), Rome, Italy ABSTRACT In the last years, an increasing attention has been devoted to seismically induced site effects, particularly punctual and linear liquefaction ones. The large availability of geotechnical investigation publicly disclosed by institutional and scientific communities, combined with the evidence collected from previous seismic events, represents an opportunity for liquefaction probability prediction through machine learning approaches. In particular, the machine learning ML regression algorithm "Ensemble Bagged Tree" allows to predict the probability of liquefaction occurrence. The geotechnical investigations provided by Geyin et al., 2021 for Canterbury (New Zealand) has been used as learning site since it is one of the most comprehensive geotechnical datasets freely available. This model is based on the correlation of liquefaction occurrence with the main features controlling the phenomenon triggering. It proved to be capable to predict the occurrence of liquefaction with good accuracy in training phase. This trained model will be tested to predict liquefaction probability for national and international case studies. References Geyin M., Maurer B.W., Bradley B.A., Green R.A. and van Ballegooy S.; 2021: CPT-based liquefaction case histories compiled from three earthquakes in Canterbury, New Zealand. Earthquake Spectra, 37(4), doi: 10.1177/8755293021996 Contact author: chiara.varone@igag.cnr.it