Explainable Machine-Learning Predictions for Peak Ground Acceleration

Peak ground acceleration (PGA) prediction is of great significance in the seismic design engineering structures. Machine learning a new method to predict PGA and does have some advantages. To establish explainable models PGA, 3104 sets uphole downhole records collected by KiK-net Japan were used. The feature combinations that make perform best selected through selection. peak bedrock (PBA), predominant frequency (FP), depth soil when shear wave velocity reaches 800 m/s (D800), (Bedrock Vs) used as inputs PGA. XGBoost (eXtreme Gradient Boosting), random forest, decision tree established, results compared with numerical simulation influence between input features model analyzed SHAP (SHapley Additive exPlanations) value. show R2 training dataset testing reach up 0.945 0.915, respectively. On different site classifications intervals, are better than forest model. Even if non-integrated algorithm (decision model) used, its effect methods. values three machine same distribution densities, each on consistent existing empirical data, which shows rationality provides reliable support for results.