Estimation of economic seismic loss of steel moment-frame buildings using a machine learning algorithm

Abstract

In this study, the effect of modeling-related uncertainties on the expected annual losses of modern code-compliant steel moment-frame buildings is analyzed. Probabilistic structural models are initially employed to account for all the critical sources of uncertainty. Then, these structural models are used to develop machine-learning-based prediction models to estimate the expected annual losses and the associated economic contributors; the developed machine-learning-based prediction models exhibit an excellent performance in the prediction of the economic seismic losses of steel frame buildings. The effect of structural-modeling-related uncertainties on each loss contributor is also evaluated; the effect of uncertain modeling parameters is observed to be more pronounced on loss contributors such as demolition and structural collapse losses that are controlled primarily by ground motions with a low probability of earthquake occurrence.