Machine learning-based model for the ultimate strength of circular concrete-filled fiber-reinforced polymer–steel composite tube columns

Abstract

This study introduces a machine learning (ML)-based model for predicting the ultimate strength of circular concrete-filled fiber-reinforced polymer (FRP)–steel composite tube (CFSCT) columns. The support vector regression (SVR), back propagation neural network (BPNN), and random forest (RF) ML algorithms were trained and tested based on an extensive dataset containing 305 data samples of circular CFSCT columns under axial loads from existing literature. Seven parameters were selected as input variables, and the ultimate load of the circular CFSCT columns under an axial load was selected as the output variable to develop the ML-based models. The ML-based models were evaluated by comparing them with four existing empirical models previously developed by researchers. The results indicate that the ML-based models are good alternatives to existing empirical models, considering their superior prediction accuracy and applicability. The coefficient of determination (R2) of the SVR model was 0.992, demonstrating better performance than the BPNN and RF models with R2 of 0.984 and 0.982, respectively. However, only the SVR and BPNN model are considered applicable for the ultimate load prediction since the RF model exhibited an overfitting problem. Moreover, the results of the comparison between the ML-based and empirical models were visualized to illustrate the superiority of the ML-based models over existing empirical models in prediction accuracy and dispersion. Finally, sensitivity and parametric analyses were performed to analyze the influence of the input variables on the output variable. The results indicate that concrete compressive strength, steel tube thickness, and FRP thickness positively affect the ultimate strength of CFSCT columns. This study can provide a reference for the design of CFSCT columns.