Predicting the Strength of Eccentrically Compressed Short Circular Concrete Filled Steel Tube Columns

The process of predicting the load-bearing capacity of eccentrically compressed circular concrete filled steel tube (CFST) columns using machine learning algorithms is investigated. The relevance of the work is established by the need to improve the accuracy of engineering calculations in the context of increasingly complex architectural solutions. The purpose of the study is to develop and evaluate the effectiveness of intelligent models for reliable prediction of CFST column strength based on key parameters of the structure and materials. The object of the study was short, eccentrically compressed CFST columns of circular cross-section. The input parameters of the machine learning models were the outer diameter of the column section, tube wall thickness, concrete strength, yield strength of steel and relative eccentricity. The load-bearing capacity of the column was taken as the output parameter. CatBoost and Random Forest Regressor (RFR) algorithms with hyperparameter optimization using the Optuna library were used for forecasting. The quality of the models was assessed using the MAE, MSE, and MAPE metrics. As a result of the study, intelligent models were developed. The CatBoost model demonstrated better accuracy rates (MAE = 67.1; MSE = 86.2; MAPE = 0.07%) compared to RFR (MAE = 72.6; MSE = 89.7; MAPE = 0.15%). The feature importance analysis showed that the outer diameter of the column and the relative eccentricity have the greatest influence on the bearing capacity. Correlation analysis confirmed the high dependence of the output parameter on these factors. The obtained results are recommended for use in calculation modules and supporting engineering systems for design solutions of load-bearing structures.