Strength estimation of concrete columns laterally confined with CFRP and TSR using various machine learning models

This study presents enhanced multilayer perceptron (MLP) neural network models—MLP-Bat and MLP-TLBO—for predicting the compressive strength of concrete columns transversely confined with carbon fiber-reinforced polymer (CFRP) sheets and transverse steel reinforcement (TSR) such as ties or spirals. A comprehensive experimental database was compiled from the literature and utilized for artificial neural network (ANN) modeling. The proposed models were benchmarked against existing empirical formulations and theoretical models. Results demonstrated that the MLP-Bat model outperformed the MLP-TLBO model and empirical formulations in terms of prediction accuracy, with statistical coefficients of MAE=6.953, RMSE=7.857, and R2=0.988 for MLP-Bat, compared to MAE=7.865, RMSE=7.391, and R2=0.981 for MLP-TLBO, and MAE=11.875, RMSE=9.323, and R2=0.948 for the empirical model. These findings highlight the superior accuracy of the MLP-Bat model, making it a reliable tool for predicting the axial strength of CFRP-confined concrete columns reinforced with TSR.