Machine Learning-Assisted Identification of Vulnerable Historic Buildings in Urban Environments

Abstract

The vulnerability assessment of architectural cultural heritage represents a challenging, however necessary, activity towards defining risk preparedness plans for historic urban environments and more resilient communities. Nevertheless, the singularities and specificities that historical constructions often present the need for customising/tailoring and adapting generalised and common vulnerability assessment approaches. The balance between detail and scale when surveying historical structures implies that the feasibility of achieving urban-scale data acquisitions depends primarily on adopting relatively simple models and descriptions. In this sense, several simplified (often parametric) approaches have been developed for assessing the seismic vulnerability of historical constructions based on a limited set of descriptors. Although the selection of the parameters that influence the seismic response of the structures have been drawn from empirical observations (e.g., in the aftermath of intense seismic events), such regressive analysis may be improved and facilitated by employing Machine Learning categorisation algorithms. This chapter investigates the usability of parameter-based screenings of historical cities for assembling an urban-scale database that is further used for assessing the analytical vulnerability of historical constructions based on existing intensity/damage models. Furthermore, the observations acquired in the aftermath of a strong seismic event (that of the 19th of September 2017 in Mexico) are herein used for calibrating the model using a Random Forest Classifier algorithm, achieving a more representative intensity/damage model and, therefore, representing an opportunity for obtaining typologically tailored seismic vulnerability models in the context of a feasible urban-scale survey and post-earthquake observations. These trained algorithms are valuable tools in supporting risk preparedness and management policies based on the results of rapid screenings for proactively identifying vulnerable assets and evaluating the impact of different mitigation measures.