Data-driven seismic mechanical performance evaluation of RC columns based on adaptive optimization ensemble learning method integrating random forest and back propagation neural network

• A novel ensemble learning model (AO-RF-BP) is developed. • A dynamic weighting and adaptive optimization strategy is proposed. • The superiority of the proposed method in seismic mechanical performance is verified. . The mechanical properties of reinforced concrete (RC) columns are key to assessing how structures perform during earthquakes. Current methods for analysis lack sufficient accuracy and efficiency in calculations. The existing analysis methods have shortcomings in calculation accuracy and efficiency. To address this, this study proposes an adaptive optimization ensemble learning model (AO-RF-BP) for seismic mechanical performance evaluation. The model integrates the strengths of random forest (RF) and back propagation neural network (BP) models, employing the dynamic weighting strategy based on mean absolute error (MAE). An improved black-winged kite optimization algorithm (IBKA), based on Levy flight and adaptive t-distribution strategies, is used to optimize the hyperparameters of the base models. Furthermore, to enhance the interpretability of prediction results, the SHAP method is employed for global model interpretation, analyzing the impact of input features (e.g., geometric parameters, material properties, and load characteristics) on seismic mechanical performance. Based on the comparative study of 1101 sets of RC column test data, the model demonstrates high accuracy and generalization ability in evaluating critical seismic mechanical performance. The research results provide a new idea for the efficient evaluation of the seismic performance of RC columns and expand the application potential of ensemble learning in the performance prediction of complex structures.