Research Information System
JOURNAL OF EARTHQUAKE ENGINEERING, 2025 (SCI-Expanded)
The February 6, 2023, doublet earthquake in Kahramanmara & scedil;, T & uuml;rkiye, known for its multi-fault rupture, caused widespread destruction and affected numerous buildings across 17 provinces. This research focused on the utilization of machine learning (ML) models for the classification of an extensive earthquake-induced building damage dataset collected from detailed post-disaster inspections of 2,432,871 buildings, along with seismic activity data from nearby stations. The objective was to construct an ML-based model capable of accurately predicting the severity of earthquake-induced damage to buildings. The proposed model integrated the general characteristics of 965,270 reinforced concrete (RC) buildings to address the critical need for rapid and precise damage assessment following seismic events. Focusing on the robustness and reliability essential in complex seismic scenarios, this study evaluated nine ML models, including Decision Trees, Random Forest, XGBoost, and Logistic Regression. The evaluation of the performance metrics demonstrated the high predictive performance of Random Forest, achieving an accuracy of 93%. This model excelled in overall accuracy and performed comparatively better in the prediction of collapse instances. While KNN also achieved a similar accuracy and F1 score, it demonstrated lower performance across other metrics and faced limitations in accurately predicting building collapses. The results also demonstrated that using interpolated PGA data with the KDTree method can enhance the responsiveness of damage detection systems in the immediate aftermath of an earthquake. The findings provide insights into the optimization of model training within complex and large-scale datasets and establish the potential of ML models to enhance the efficiency and precision of post-seismic structural damage assessment.