Machine learning approaches for seismic vulnerability assessment of urban buildings: A comparative study with analytic hierarchy process

Implementing pre-disaster earthquake strategies is essential for minimizing post-earthquake impacts. In this vein, one key strategy is to assess the seismic vulnerability of existing urban buildings, enabling the adoption of necessary rehabilitation procedures. Here, important parameters influencing the seismic vulnerability of urban buildings are first documented and prioritized then using multi-criteria decision-making tools. This results in a vulnerability index (VI) representing the potential earthquake damage. Using semi-supervised machine learning (ML) methods, the corresponding VI is determined, and the results are compared with different methods. Various ML-based methods are analyzed for the available dataset to identify the most effective approach for this study. This methodology is then applied to an urban region to assess the VI not only for the current year (i.e., 2024) but also to predict it for 2044 and 2064. The VI of buildings indicates that approximately 60 % and 90 % of the structures in the studied region will experience significant damage to earthquakes in the years 2044 and 2064, respectively. In the final step, various ML methods are evaluated for data classification. Decision tree and random forest methods achieve an accuracy of over 95 %, while linear regression is utilized for predicting the index value, resulting in an R-squared error rate of approximately 91 %. • An ML-based index for seismic vulnerability assessment of urban buildings is developed. • The results of the ML-based index are compared with those found in conventional Analytic Hierarchy Processes. • The developed index forecasts the seismic vulnerability status of the cases studied 20 and 40 years later. • Various ML methods are evaluated for data classification to understand which ML algorithms are the most effective ones for predicting earthquake vulnerability in urban buildings.