Machine learning (ML) algorithms for seismic vulnerability assessment of school buildings in high-intensity seismic zones

Abstract

Ensuring seismic resilience of school buildings is crucial for safeguarding their occupants during earthquakes. This paper focuses on assessing the seismic vulnerability of school buildings constructed in the Kashmir region of Pakistan after the 2005 earthquake, which claimed the lives of 19,000 school-going children. It explores the feasibility of utilizing machine learning (ML) algorithms for enhanced rapid screening of schools to establish fragility information. The study is based on data collected in the Kashmir region and focuses on assessing representative reinforced concrete (RC) and unreinforced masonry (URM) school buildings. To determine structural fragility curves, Incremental Dynamic Analyses (IDA) are performed, simulating fifteen historical earthquakes. Four different ML models are investigated to predict fragility curves, including Random Forest (RF), Artificial Neural Networks (ANNs), Extreme Gradient Boosting (XGBoost), and Extremely Randomized Tree Regressor (ERTR). The performance of the algorithms is compared using performance metrics such as precision, accuracy, and f1 score. The study identified XGBoost and RF as the highest performing algorithms, achieving highly satisfactory accuracy with the correlation coefficients of 0.91 and 0.81 for RC schools, and 0.88 and 0.83 for URM schools during testing phases. Alternatively, ERTR’s performance could not justify its use for structural seismic vulnerability assessments. This highlights the significant potential of using ML algorithms for automated seismic vulnerability evaluation of buildings, greatly reducing the overall computational burden while maintaining high accuracy and reliability.