Random Forest based estimate to assess the damages of future earthquakes: preliminary results

In this paper we present a case study where the Random Forest (RF) Classifier, has been used to estimate the damage to buildings caused by a (possible) future earthquake, starting from the data of past earthquakes. This preliminary work is based on the Shakedado dataset, containing information on buildings and ground shaking parameters for the six major earthquakes that occurred in Italy between 1981 and 2012. We perform the following two conceptual experiments: E1. Assume that Emilia seismic sequence has just ended and the data from the other major earthquakes that have occurred in the past (L'Aquila, Pollino and Irpinia) are available. We calculate the damage level for each building in the Emilia dataset. E2. Assume that the Pollino seismic sequence has just ended and the data from the other major earthquakes with comparable magnitude (L'Aquila, Emilia) are available. We calculate the damage level for each building in the Pollino dataset. Both training and test datasets contain only masonry buildings located within 10 km of the main shock of each sequence. The results demonstrate the ability of the RF algorithm to discriminate between light/no and medium/severe damaged building, with a good accuracy especially for E1.