Model selection for prediction of strong ground motion peaks in Turkiye

Abstract

This study focuses on the model selection of peak ground acceleration (PGA), peak ground velocity (PGV), and peak ground displacement (PGD) prediction utilizing strong motion data of earthquakes that occurred in Turkiye. An inventive gradient boosted model (GBM) predicted the PGA, PGV, and PGD using a database obtained from The Ministry of Interior Disaster and Emergency Management Presidency of Turkiye (AFAD). Alternative classification and decision tree models like K-Star algorithm, Sequential Minimal Optimization Regression (SMOreg), Random Forest (RF), and K-Nearest Neighbors (K-NN) were utilized to make comparisons with the results obtained from GBM. Furthermore, this comparison also includes the PGA prediction outcomes obtained through the use of the artificial neural network (ANN) technique. Earthquake magnitude (Mw), epicentral distance (Repi), focal depth (FD), and mean shear-wave velocity from the surface to a depth of 30 m (Vs30) were considered as independent input parameters. Akaike information criterion (AIC) and Schwartz Bayesian criterion (SBC) were employed to evaluate the prediction performance of all models. The results obtained from the GBM model were found to be better regarding PGA prediction than the other models. These results were revealed models to be a promising approach for the prediction of PGA, and capable of representing the complex relationship between predicted and input parameters.