Applications of and considerations for using machine learning and deep learning tools in earthquake engineering, with focus on soft story building identification

Abhineet Gupta, Todd MacDonald, & Debbie Weiser

Published August 14, 2019, SCEC Contribution #9598, 2019 SCEC Annual Meeting Poster #298 (PDF)

Machine learning and deep learning technologies are being increasingly used in applied fields including earthquake engineering, with example applications including predicting damage levels from imagery or forecasting earthquakes from acoustic signals. In this study, we present one application of machine learning and deep learning in order to identify soft story buildings that are highly susceptible to collapses during earthquakes. Our two-step model is able to estimate soft story buildings with high recall, from among all buildings in a city. We discuss some of the limitations of our model. We also present that traditional machine learning metrics, like confusion matrices, are not suitable for the problem of damage estimation after an earthquake due to aleatory uncertainties, and present alternative approaches for evaluating these models.

Key Words
machine learning, soft story, deep learning, metrics

Citation
Gupta, A., MacDonald, T., & Weiser, D. (2019, 08). Applications of and considerations for using machine learning and deep learning tools in earthquake engineering, with focus on soft story building identification. Poster Presentation at 2019 SCEC Annual Meeting.

Related Projects & Working Groups
Computational Science (CS)