Seismic Rocking Response Classification Through the Lens of a Machine Learning Methodology

This paper explores the application of machine learning methods to the classification of the response of rocking structures when subjected to recorded earthquakes. Specifically, this study focuses on efficiently predicting whether a block, after commencing rocking motion, overturns (i.e. collapses) or undergoes safe rocking motion (i.e. returns to the initial rest position). To this end, this research adopts random forest algorithms to classify the pure rocking response of a rigid block. It also aims at identifying the main seismic characteristics, in the form of intensity measures (IMs), that govern such classification. This study offers a random forest model able to classify with great accuracy the response of a rigid rocking block when subjected to recorded earthquakes. Importantly, the results highlight the importance of velocity characteristics of the seismic waveform on the overturning mode. Finally, this work recommends that intensity-based and frequency-based IMs may serve as an efficient link between seismic signal and overturning.