Prediction of shear contribution of fibre-reinforced polymer-strengthened reinforced concrete beams using machine learning

Machine learning (ML) techniques are widely used for reliable forecasting of the shear strength contribution of fibre-reinforced polymer (FRP)-strengthened reinforced concrete beams. However, the shear behaviour of beams strengthened by FRP that have low shear span-to-effective depth ratios (av/d), as well as the interaction between vertical steel shear reinforcement and FRP strips, cannot be well explained by the research that is currently available in the literature. This study provides five supervised ML algorithms (ridge regression, random forest, support vector regression, extreme gradient boosting and categorical gradient boosting (CatBoost)) to address this gap. A comparative analysis was conducted using three existing analytical models and three established standard guidelines to prove the reliability of ML models. Grubbs’ test with median imputation was used for outlier detection, followed by robust scaling to ensure data consistency. A Shapley additive explanations-based feature importance plot was implemented to interpret the influence of individual parameters on the FRP shear contribution (Vf). The results demonstrated that CatBoost achieved the highest accuracy levels for FRP shear-contribution predictions (R = 0.96; mean absolute error = 13.70 kN; mean absolute percentage error = 18.07%; root mean square error = 21.80 kN; and Nash–Sutcliffe efficiency = 0.92).