DATA-DRIVEN PREDICTION OF SEISMIC INTENSITY DISTRIBUTIONS FEATURING HYBRID CLASSIFICATION-REGRESSION MODELS

Earthquakes are among the most immediate and deadly natural disasters faced by humans. Accurate prediction of the extent of earthquake damage and assessment of potential risks can save numerous lives. In this study, we developed a hybrid model combining classification and regression models, capable of predicting seismic intensity distributions based on the following earthquake parameters: location, depth, and magnitude. As these models are completely data-driven, they can predict seismic intensity distributions without geographic information. The dataset comprises seismic intensity data from earthquakes that occurred in the vicinity of Japan between 1997 and 2020. It includes 1,857 instances of seismic intensity data for earthquakes with a magnitude of 5.0 or greater, sourced from the Japan Meteorological Agency. Regression and classification models were trained, then combined to take advantage of each other and create a hybrid model. The proposed model outperformed commonly used ground-motion prediction equations (GMPEs) in terms of the correlation coefficient, F1 score, and MCC. Furthermore, the proposed model can predict abnormal seismic intensity distributions, a task that conventional GMPEs often struggle to achieve.