Prediction of Effective Duration of Vertical Ground Motions Based on Machine Learning Algorithms

Vertical ground motions during an earthquake refer to the upward and downward movements of the ground caused by seismic waves. These vertical motions can be caused by several factors, including the type of fault that generated the earthquake, the depth of the focus of the earthquake, and the type of soil or rock at the surface. Vertical ground motions are often neglected by the engineering community; however, they can have significant impacts on buildings, particularly tall structures, and can contribute to damage during an earthquake. The objective of this paper is to forecast the effective duration of vertical ground motions using machine learning techniques. In this regard, 100 near-fault pulse-like vertical ground motions collected across the globe are characterized to identify the key ground motion intensity measures. Based on the engineering feature analysis, it was found that magnitude, epicentral distance, frequency, and PGA were selected as input variables. The results indicate that tree-based model was found to be promising for forecasting the effective duration of ground motions.