A machine learning-based framework for seismic vulnerability assessment of reinforced concrete educational facilities

Abstract

This study presents the development of a machine learning (ML) based framework for assessing the seismic vulnerability of existing educational reinforced concrete (RC) buildings under the jurisdiction of the Rajdhani Unnayan Kartripakkha (RAJUK) in Dhaka. The conventional three major stages of assessment methods are often resource-intensive and time-consuming, especially when applied to large building stocks. The primary objective is to assess the seismic vulnerability of existing RC educational buildings in similar contexts using the ML method, focusing in predicting analytical parameters Story Shear Ratio (SSR) as a critical analytical risk indicator, implementing Rapid Visual Assessment (RVA) 8 parameters. The RVA parameters are construction year, building condition, number of stories, typical floor area, redundancy, pounding, plan irregularity and elevation irregularity, and corresponding building’s SSR value in the preliminary Engineering Assessment (PEA) survey. Three well-known ML models, including Support Vector Regression (SVR), Random Forest Regression (RFR), and Artificial Neural Networks (ANN), were employed to predict SSR using RVA parameters. The dataset of 268 RC educational buildings was collected from the RAJUK. Based on the analysis, the SVR model obtained a higher coefficient of determination (R²) of 0.34 than the 0.17 and 0.16 of the RFR, and ANN models and 0.038, 0.04, and 0.04 for the Mean Square Error, respectively, though all models exhibited limited explanatory power for SSR. The findings reveals that the SVR handled comparatively well the complexities and nonlinearities in the dataset. This study proposes a cost-effective ML framework for seismic vulnerability assessment, with potential to support urban resilience efforts following further validation.