Investigation of Joint Types on the Seismic Performance Prediction of RC Beam-Column Joints Based on Machine Learning Method

Recently, data intelligence (DI) models based on machine learning have been extensively applied to the study of reinforced concrete (RC) beam – column joints. However, the datasets used in most DI models typically combine both interior and exterior joints while the joint type is a complex variable that has received limited attention in previous research. In this study, a dataset comprising 310 reinforced concrete beam – column joints subjected to low-cycle loading was compiled, including 171 interior joints and 139 exterior joints. To examine the influence of joint types on the prediction of seismic performance (i.e. failure modes and peak shear stress), three separate datasets were constructed: one containing only interior joints (IJ), one with only exterior joints (EJ), and one combining both types (IEJ). In addition, to balance computational efficiency and prediction accuracy, the best-first search strategy was integrated with conventional machine learning algorithms to develop an improved predictive model. The proposed model identifies the most suitable feature combinations for each algorithm in predicting joint performance and was validated against conventional machine learning algorithms with and without feature selection, as well as typical joint theory models. Corresponding DI models for IJ, EJ, and IEJ datasets were established to predict failure modes and peak shear stress of beam – column joints. Finally, feature importance indices were calculated using the selected feature sets and interpretability techniques, and the influence of joint types on prediction performance was systematically analyzed.