Rapid damage assessment and rotation angle model of EEP-HSSBC connections under post-fire earthquake using machine learning

To achieve accurate and rapid assessment of the seismic damage state of extended end-plate high-strength steel beam-column connections after fire, this study utilizes seven parameters as input variables: column width-tothickness ratio, column height-to-thickness ratio, beam width-to-thickness ratio, beam height-to-thickness ratio, beam stiffener length-to-thickness ratio, temperature-induced damage, and seismic intensity levels. The output variable is the damage state of the connections under seismic loading after fire. Eight machine learning models were employed: Category Boosting, K-Nearest Neighbors, Artificial Neural Network, Na & iuml;ve Bayes, Decision Tree, Random Forest, Adaptive Boosting, and Extreme Gradient Boosting. The results demonstrate that the Extreme Gradient Boosting model achieved the highest prediction accuracy (0.9) on the test set, followed by the Artificial Neural Network model (0.79). Among the models, the Decision Tree, Category Boosting, Artificial Neural Network, and Extreme Gradient Boosting models exhibited high recall and precision, with the Extreme Gradient Boosting model achieving the highest values for both metrics. By comparing the predicted probabilities across different models and damage states, the Extreme Gradient Boosting and Category Boosting models provided the best predictive performance, with mean absolute errors of 4.18 and 3.77, and mean squared errors of 27.83 and 23.75, respectively. An analysis of input variable importance using the Extreme Gradient Boosting model revealed that the beam stiffener length-to-thickness ratio was the most significant factor influencing postfire seismic damage prediction, with an importance coefficient of 24.3 %. This was followed by the beam heightto-thickness ratio (19.9 %) and the beam width-to-thickness ratio (17.5 %). These findings highlight the effectiveness of the Extreme Gradient Boosting model in predicting seismic damage states and provide valuable insights into the key factors affecting the post-fire performance of extended end-plate high-strength steel beamcolumn connections.