Understanding critical masonry building attributes shaping vulnerability to blast loads: Data-driven insights from the 2020 Beirut explosion

This study investigates the building attributes shaping the resiliency and vulnerability of unreinforced masonry buildings when subjected to blast loads, utilizing a dataset derived from the aftermath of the 2020 Beirut Ammonium Nitrate explosion. Employing a suite of machine learning algorithms and methodologies, including the novel combination of nested cross-validation with SMOTE and class reduction, the study identifies key building attributes that significantly impact vulnerability. Our proposed methodology achieves over 90% accuracy in predicting damage levels on unseen datasets across multiple algorithms. Contrary to expectations, the presence of soft-stories and structural modifications, while significant in other contexts, did not exhibit a significant association with damage severity, suggesting unique responses to blast loads compared to other types of loading. The developed methodology can also be applied to other disaster types to better understand the attributes affecting the behavior of URM buildings in different contexts. The results of this research can help architects, engineers, and policymakers develop targeted retrofitting strategies, refine building regulations, and create effective zoning protocols to enhance the resilience of historic urban environments against blast hazards.