Utilizing advanced machine learning approaches to assess the seismic fragility of non-engineered masonry structures

Seismic fragility assessment provides a substantial tool for assessing the seismic resilience of these buildings. However, using traditional numerical methods to derive fragility curves poses significant challenges. These methods often overlook the diverse range of buildings found in different regions, as they rely on standardized assumptions and parameters. Consequently, they may not accurately capture the seismic response of various building types. Alternatively, extensive data collection becomes essential to address this knowledge gap by understanding local construction techniques and identifying the relevant parameters. This data is crucial for developing reliable analytical approaches that can accurately derive fragility curves. To overcome these challenges, this research employs four Machine Learning (ML) techniques, namely Support Vector Regression (SVR), Stochastic Gradient Descent (SGD), Random Forest (RF), and Linear Regression (LR), to derive fragility curves for probability of collapse in terms of Peak Ground Acceleration (PGA). To achieve the research objective, a comprehensive input/output dataset consisting of on-site data collected from 646 masonry walls in Malawi is used. Adopted ML models are trained and tested using the entire dataset and then again using only the most highly correlated features. The study includes a comparative analysis of the efficiency and accuracy of each ML approach and the influence of the data used in the analyses. Random Forest (RF) technique emerges as the most efficient ML approach for deriving fragility curves for the surveyed dataset in terms of achieved lowest values for evaluation metrics of the ML methods. This technique scored the lowest Mean Absolute Percentage Error (MAPE) of 16.8 %, and the lowest Root Mean Square Error (RMSE) of 0.0547. These results highlight the potential of ML techniques, particularly RF, in derivation of fragility curves with proper levels of accuracy.