Machine Learning Prediction of the Compressive Bearing Capacity of Concrete-Filled Steel Tubes Using Random Forest

Concrete-filled steel tube (CFST) members are widely used in long-span and high-rise structures due to their high load-bearing capacity and structural efficiency. Accurate prediction of their compressive bearing capacity is essential for reliable design. In this study, a data-driven prediction model based on the Random Forest (RF) algorithm was developed using a database of 154 axial compression tests. A total of 24 parameters, including geometric dimensions, material properties, and sectional characteristics, were considered as input variables, and the model was optimized through five-fold cross-validation and hyperparameter tuning. The results indicate that the proposed model achieves high accuracy and stability, with mean predicted-to-experimental ratios of 1.002 and 0.989 for the training and testing sets, respectively, and maximum deviations within 15%. Compared with existing design codes and alternative machine learning methods, the RF model improves prediction accuracy by approximately 9% and exhibits strong generalization capability. Furthermore, independent experimental validation using nine CFST column tests confirms its reliability, with prediction errors within 5%. These findings demonstrate that the proposed model provides an effective and practical tool for predicting the compressive bearing capacity of CFST members in engineering applications.