A stacking-CRRL fusion model for predicting the bearing capacity of a steel-reinforced concrete column constrained by carbon fiber-reinforced polymer

In a two-level stacking algorithm framework, a fusion model (stacking-CRRL) of categorical boosting (Catboost), random forest regression (RFR), ridge regression (RR), and Least absolute shrinkage and selection operator (LASSO) is proposed and shown to accurately predict the load capacity in axial compression of steel-reinforced concrete columns (SRCCs) clad in carbon fiber-reinforced polymer (CFRP). Sparse initial data were extended by synthetic minority oversampling in the model-building process, and 12 model input features were identified after eliminating redundant features using Spearman correlation coefficients. The prediction performance of five boosting models, two bagging models, and three traditional machine learning (ML) models were compared. The Catboost, RFR, and RR models were selected as the base learners, and LASSO regression was chosen for the meta-learner. The prediction performance of different algorithmic models before and after synthetic minority oversampling technique (SMOTE) processing is compared, and the stacking-CRRL fusion model established is compared with that of established prediction techniques. The Shapley additive explanations technique was applied and discussed the impact of input features on the bearing capacity of SRCCs. The results demonstrate that the prediction performance of the proposed stacking-CRRL fusion model surpasses that of the alternative models tested, that of a published prediction equation, and that of an Abaqus simulation.