A data-driven approach for rapid seismic risk prediction of RC buildings

Abstract In regions with high seismic activity, the rapid and accurate determination of vulnerable buildings within the existing building stock is a crucial step toward ensuring life safety by enabling preventive measures before future earthquakes. This study presents an artificial intelligence (AI)-based methodology for the rapid and reliable estimation of seismic risk levels in reinforced concrete (RC) frame structures. In this context, the existing building stock was simulated by incorporating parameters such as inadequate material strength, workmanship defects, and usage-related defects in load-bearing elements, which are commonly observed in structures. To establish a robust framework, nonlinear performance analyses were conducted in accordance with the seismic code, and seismic risk levels were classified based on performance outcomes. Through this simulation, 1,680 building profiles were generated, and their corresponding risk levels were determined. The resulting dataset was used to train and evaluate eight different machine learning (ML) models with hyperparameter optimization. Among these, the XGBoost model achieved the highest predictive accuracy of 92% on the test dataset. The results indicate that AI models can rapidly and accurately predict the seismic risk levels of RC buildings, providing a reliable basis for large-scale risk assessment and pre-earthquake decision-making.