Machine learning model for building seismic peak roof drift ratio assessment

The peak roof drift ratio is one of the most important engineeringparameters to describe the expected seismic damage in a building. A predictive model of the drift ratio was developed using amachine learning approach (Gaussian process regression model)on a dataset of approximately 11,800 records from 34 monitoredbuildings in Japan. Four predictors for ground motion and threepredictors for building vulnerability are used in the machinelearning modelling. The residual analysis shows a reduction of50% compared to the state of the art. The Gaussian processregression model is applied in a second analysis on an originaldataset of approximately 4,500 records for 127 monitored buildings in Italy. A satisfactory comparison emerges by comparing thedrift ratio prediction map with the observed damage pattern produced by satellite imagery for a test site in central Italy after the2009 earthquake. The drift ratio map plays an important role inthe simulation of an earthquake scenario at regional scale, whichis needed by Civil Protection for emergency planning and management activities.