Development of a deep learning‐based rapid visual screening method for seismic vulnerability assessment of existing <scp>RC</scp> buildings in Turkey

Abstract Urbanization and population growth have increased the existing building stock, making it more challenging to assess the seismic safety of buildings due to time constraints, a lack of skilled personnel, and high economic costs. In this study, a rapid visual screening method (RVS) was utilized to assess and prioritize the seismic vulnerability of reinforced concrete (RC) buildings. Accordingly, an integrated model combining deep feature residual networks, a deep learning‐based architecture relying on residual blocks, and XGBoost was proposed. Additionally, the five most influential parameters for determining the seismic vulnerability of buildings were identified using the XGBoost technique. RVS methods were used to collect data on RC buildings following earthquakes in Afyon, Bingöl, Van, and Kahramanmaraş, resulting in a dataset of 372 structures. The model's performance was evaluated using accuracy, precision, recall, F1‐score, specificity, and AUC metrics. The proposed architecture achieved a high accuracy rate of 94.66% in determining the seismic vulnerability of buildings. Furthermore, using only the five most critical features, an accuracy of 82.66% was obtained. Sensitivity analysis was performed to see the effect of the parameters used on the model. In addition, the stability of the model was tested against parameter changes or possible erroneous inputs. The results indicated that although the model was sensitive to parameter changes, its predictions remained within certain limits and showed a stable behavior against possible errors.