Development of seismic fragility curves for RC/MR frames using machine learning methods

Reduction of earthquakes fatality requires a study on seismic risk estimation in various sites. Deriving seismic fragility curves of structures make it easier to investigate the seismic vulnerability of structures. One of the new methods in estimating the response of structures is the use of machine learning techniques. The machine learning process is the statistical inference of identifying patterns in data sets to make decisions for new cases. In this paper, deriving seismic fragility curve for a reinforced concrete (RC) moment resisting (MR) frame using machine learning methods is done. Six different classification-based methods, including K-nearest neighbors, decision tree, naïve Bayes, logistic regression, linear regression, and ridge regression, are explored on two data sets to build machine learning models. The machine learning algorithms used in this study take the output from nonlinear analysis of the studied frame as its input. Ground motion selection is based on Baker's proposed ground motion records and the generalized conditional intensity measures (GCIM) methodology. The fragility curve of assumed frame has also been determined using the multiple stripes analysis (MSA) method, and the difference between these two curves has been compared. The results show that the difference between the fragility curves determined based on MSA and machine learning methods is less than 10%. Also, the distribution of earthquake ground motion records at different seismic intensities is effective in the performance of machine learning models, so that if the records in low and medium intensity levels are much more than the high intensity level, the model tends to predict the non-collapse of the structure. © 2022, The Author(s), under exclusive licence to Springer Nature Switzerland AG.