Modelling the Importance of Ground and Strong-Motion Variables on the Damage Status in the 2023 Kahramanmaraş Earthquakes Using Supervised Machine Learning

Abstract

The damage status of 44 locations was investigated, incorporating ground condition parameters such as Vs30, engineering bedrock depth (EBd), and predominant frequency (f0), as well as strong-motion parameters including PGA, Repi, and Rrup (epicentre and rupture distance, respectively). Various machine learning methods—logistic regression (LR), classification and regression trees (CART), random forest (RF), support vector machine (SVM), k-nearest neighbours (KNN), and artificial neural networks (ANN)—were employed to evaluate the dataset through three approaches: the complete parameter set, solely ground-based parameters, and strong-motion parameters alone. Model performance, measured by Area Under the Curve of the Receiver Operating Characteristic (AUC-ROC), ranged from 0.466 to 0.989, with KNN achieving the highest performance (0.989) when using the complete dataset and 0.988 with ground-based parameters alone. The analysis highlighted EBd and f0 as the most significant contributors to damage outcomes (normalised variable importance of 100% and 85%, respectively), demonstrating strong correlations with structural vulnerability. Among earthquake-related parameters, PGA was identified as the most influential factor in models established through strong-motion parameters, whereas Repi and Rrup demonstrated a considerably lower influence. On the other hand, specificity values (determining no-damage status) consistently exceeded sensitivity (determining damage status) in models based solely on earthquake parameters. Overall, the outputs demonstrate that traditional seismic hazard approaches based on earthquake parameters could provide a broad framework for risk mitigation; local site conditions, particularly EBd and f0, are the primary drivers of seismic risk. Integrating these detailed ground parameters into seismic assessments is critical for improving predictive accuracy and advancing earthquake engineering practices.