2022 Volume 22 Issue 4 Pages 4_23-4_38
We constructed ground motion evaluation models of maximum accelerations and response spectra by supervised machine learning based on the strong motion database. The normal logarithmic standard deviations of the ratios of the predicted values to the observed ones in our models are 0.18-0.21, which is less than the variation of the previous ground motion prediction equations. The generalizability of the model was tested using three additional test earthquakes that occurred later than the training dataset. The results showed that the prediction accuracy decreased for earthquakes with features that were not included in the training dataset, but the model with features based on prediction results by the previous ground motion prediction equation could compensate for the bias and lack of training data.
