Predicting unbalanced moment and drift ratio in RC slab-column connections: A hybrid MGGP and ensemble learning approach

The seismic performance of interior slab-column connections is a critical design concern due to their vulnerability to drift-induced punching shear, particularly in flat-plate systems without shear reinforcement. Existing design codes and analytical models show limited accuracy and considerable scatter when predicting unbalanced moment capacity and deformation limits, especially for both steel- and fiber-reinforced polymer (FRP) reinforced concrete systems. This study proposes a unified data-driven framework to predict the ultimate unbalanced moment capacity (Mu) and maximum drift ratio (θmax) of interior slab-column connections without shear reinforcement. A database of 128 experimental tests from 36 published studies was compiled, covering primarily steel-reinforced specimens, with FRP-reinforced connections included to assess cross-material generalizability via the normalized Reinforcement Index. A normalized Reinforcement Index (RI) was adopted to ensure consistent material representation. Transparent closed-form equations were derived using Multi-Gene Genetic Programming, while Random Forest and Least-Squares Boosting models were developed to enhance predictive accuracy. The proposed models significantly outperform current design provisions, achieving testing R2 values of 0.949 for Mu and 0.870 for θmax. Model interpretability analysis confirmed consistency with governing mechanical mechanisms, and graphical user interfaces were developed to support practical engineering application.