A Novel Rapid Visual Screening Score Sheet for Vulnerability Assessment of Low-to-Midrise Buildings Using Machine Learning Based on the 2015 Nepal Earthquake

Rapid visual screening (RVS) is one of the quickest methods for assessing the seismic vulnerability of a large number of buildings. The conventional RVS developed by FEMA P-154 is employed to predict the damage that may occur in a building after an earthquake. However, previous studies have shown that the conventional RVS score sheet shows an accuracy of less than 30%. Further, the reliability of the conventional RVS is limited due to the absence of important parameters such as the age of the buildings, peak ground acceleration (PGA), and foundation type. In this research, a novel RVS score sheet is developed considering real building damages of the 2015 Nepal earthquake using machine learning. Four supervised machine learning models, namely, random forest, logistic regression, XGBoost, and KNN, are trained on the postearthquake building dataset of the Nepal earthquake. The developed machine learning model achieves an accuracy between 73% and 78%. Moreover, Shapley additive explanations analysis shows that the PGA, age of the buildings, and foundation type are the most influential parameters in determining seismic vulnerability. Further, the proposed RVS score sheet is used to assess certain buildings in Durgapur and Bhubaneshwar city, India. It is observed that the proposed score sheet outperforms the conventional score sheets. The developed score sheet is an efficient decision-making tool that equips authorities with a reliable and computationally independent solution for earthquake disaster mitigation.