Machine Learning for Enhanced Seismic Damage Assessment of Buildings Kourehpaz, P.; Molina Hutt, Carlos
Adopting policies for seismic risk mitigation and enhancing community resilience is conditioned on accurate seismic risk assessments. This article proposes a machine learning based framework to estimate a building’s post-earthquake damage state using ground motion intensity measures and structural properties as model inputs. The machine learning algorithms evaluated, namely, Logistic Regression, K-Nearest Neighbors, Decision Tree, Random Forest, AdaBoost, and Gradient Boosting are trained using a dataset of nonlinear response history analysis results from 36 detailed modern reinforced concrete shear wall building structural models, ranging from four to 24- stories, subjected to ~500 empirical and simulated ground motion records with a range of shaking intensities. The results suggest that the Gradient Boosting technique is the most efficient algorithm by achieving a prediction success (F1-score) of 87%. The proposed framework is also retrained to identify collapse instances, by leveraging synthetic data samples, considering multiple collapse thresholds. By means of synthetic data, on average, the Gradient Boosting technique increased the percentage of observed collapse cases that are correctly classified from 76% to 93%.
Item Citations and Data
Attribution-NonCommercial-NoDerivatives 4.0 International