A Machine Learning Framework for Estimating Fragility Curves of Low- to Mid-Rise RC Buildings
Buildings, vol.16, no.11, 2026 (SCI-Expanded, Scopus)
- Publication Type: Article / Article
- Volume: 16 Issue: 11
- Publication Date: 2026
- Doi Number: 10.3390/buildings16112127
- Journal Name: Buildings
- Journal Indexes: Science Citation Index Expanded (SCI-EXPANDED), Scopus, Applied Science & Technology Source, Avery, Compendex, INSPEC, Directory of Open Access Journals, Natural Science Collection (ProQuest), Materials Science & Engineering Collection (ProQuest), Technology Collection (ProQuest)
- Keywords: fragility curves, incremental dynamic analysis, machine learning algorithm, spectral displacement and acceleration, standard deviation and median
- Hacettepe University Affiliated: Yes
Abstract
In performance-based earthquake engineering (PBEE), fragility curves hold significant importance for a reliable risk assessment of the existing reinforced concrete (RC) structures. Fragility curves require numerous incremental nonlinear dynamic analyses, which are highly time-consuming and computationally intensive. However, predicting the fragility curve parameters of RC structures by a machine learning algorithm could effectively reduce this cost. In this study, machine learning (ML)-based numerical analyses were performed in order to predict the fragility curve parameters of the existing RC structures, considering rapidly observable structural parameters by street survey. The construction date, story number, plan irregularities, soft story, and damage states are the main variables that are considered in this study. Hence, a dataset comprising the results of 620 structural fragility analyses was compiled from the existing literature. Key fragility parameters, namely the median and standard deviation, are predicted using several machine learning algorithms, including Random Forest (RF), Extreme Gradient Boosting (XGBoost), Support Vector Machine (SVM), and Artificial Neural Networks (ANNs). The performance of the proposed models is evaluated using R2, RMSE, and MAE metrics under a five-fold cross-validation scheme. Furthermore, nonlinear dynamic analyses are conducted on a representative set of structural models to validate the machine learning predictions. The results indicate that the ANN model achieves the highest predictive accuracy, followed by ensemble tree-based methods, demonstrating the capability of machine learning approaches to effectively capture complex nonlinear relationships between seismic input parameters and structural response. The proposed framework significantly reduces computational effort while maintaining reliable prediction accuracy, offering an efficient tool for seismic risk assessment and fragility estimation of existing structures.