Machine learning for earthquake engineering analysis: Comparing regression models to predict peak ground acceleration

Advancements in machine learning have introduced powerful tools for enhancing seismic hazard assessment, offering improved predictive capabilities compared to traditional regression models. This study leverages machine learning algorithms to develop data-driven ground-motion models (GMMs) for predicting peak ground acceleration (PGA), a key parameter in seismic hazard analysis. Both parametric and nonparametric regression techniques, including linear regression, polynomial regression with second-degree terms, decision tree, and random forest, are employed. The models are trained on a comprehensive dataset comprising over 10,000 ground-motion records from small-to-moderate earthquakes (magnitude 3.5 to 5.8) with hypocentral distances up to 200 km. Predictor variables such as moment magnitude (Mw), hypocentral distance (Hypo-D), average shear wave velocity in the upper 30 meters (VS30), and focal depth (Ztor) are utilized to capture the complex relationship. Performance evaluation reveals that the random forest model significantly outperforms traditional regression-based GMMs like linear regression, demonstrating its potential to enhance seismic hazard assessment, particularly for regions prone to similar earthquakes.