Multiple Regression and Neural Network-Based Models for the Prediction of the Ultimate Strength of CFRP-Confined Columns

Carbon Fiber-Reinforced Polymers (CFRPs) are gaining popularity as a reliable strengthening technique for reinforced concrete (RC) columns. Several efficient models were developed to predict the stress–strain (σ-ε) curve of CFRP-confined concrete based on experiment findings. The ultimate strength is a crucial parameter for accurate (σ-ε) behavior prediction, since it constitutes an initial step in estimating the corresponding axial strain, as it provides a direct indication of the desired increase in strength. Literature analytical models often produce inconsistent results due to errors in estimating the confinement pressure or effectively confined area or the lack of a strong and stable correlation between ultimate strength and confinement parameters. This study looked at a large collection of experimental results from existing research. It used a statistical method (Pearson’s coefficient) to see how well ultimate strength correlated with various confinement factors. For normal-strength concrete columns with circular sections, there was a strong linear correlation between ultimate strength and the thickness of the CFRP jacket. This correlation weakened for high-strength concrete (HSC) and for rectangular columns. A sensitivity analysis was performed to identify the most influential confinement parameters, showing that the number of CFRP layers (n × t) is the most dominant factor, particularly with normal-strength concrete (NSC) in circular columns, accounting for the vast majority of the variance in ultimate strength. Using multiple linear regression equations to predict ultimate strength was also explored; this method demonstrated the best performance with HSC in circular sections, but the results were less promising with NSC. Artificial Neural Networks (ANNs) were developed and trained on the built database, and four statistical metrics were computed for evaluation (R2, RMSE, MAE, MRAE), proving highly accurate and superior to linear regression equations, with mean relative absolute errors MRAEs between 2.4–7.2% for ultimate strength prediction, opening new avenues for optimizing CFRP-strengthened element designs.