In order to accurately predict the axial compression bearing capacity of steel reinforced concrete columns (SRCC) confined by carbon fiber reinforced polymer (CFRP) under multiple influencing factors. A multivariate algorithm fusion prediction model based on Random Forest (RF), categorical boosting (Catboost), eXtreme Gradient Boosting (XGBoost) and gradient boosting regression tree (GBRT) is proposed. Firstly, the synthetic minority oversampling technique (SMOTE) algorithm is used to expand the original data set, and 10 kinds of traditional machine learning and ensemble learning model tests are carried out. Four ensemble learning models of RF, Catboost, XGBoost and GBRT with determination coefficients (R²) greater than 0.92 are selected. The hyperparameters are optimized by random search, and then the RF-Catboost-XGBoost-GBRT prediction model is formed by fusion. The bearing capacity of CFRP confined steel reinforced concrete rectangular columns is predicted. The results show that the RF-Catboost-XGBoost-GBRT model shows the best prediction performance under the two datasets, and after the original dataset is processed by SMOTE, the R² of the five prediction models is increased by an average of 20.43%, among which the R² of the RF-Catboost-XGBoost-GBRT model reaches 0.942, and the prediction error is within ±10%.