Modeling of attenuation relations with artificial intelligence methods

This study directly focused on estimation of peak ground acceleration (PGA) using strong motion data of earthquakes occurred in Turkey. The data gathered from database of Ministry of Interior Disaster and Emergency Management Presidency of Turkey (AFAD). For prediction of the PGA, Random Forest (RF), M5P tree regression model, Sequential Minimal Optimization Regression (SMOREG), Gradient Boosting Modeling (GBM), Kstar and KNearest Neighbors (KNN) methods were used. In these five methods the inputs were depth of earthquake, magnitude of earthquake, repi and v30 value of soil. The prediction performance of methods was compared by some statistical criteria such as correlation coefficient (R), mean absolute error (MAE), root mean squared error (RMSE) etc. The estimation conducted by the RF method were found to be better than the other methods. The R value of RF method was 0.97, MAE and RMSE values were 7.65 and 19.40 respectively. The results revealed that the models are a fairly promising approach for the prediction of PGA and capable of representing the complex relationship between PGA and input parameters.