Metaheuristic algorithms-optimized machine learning models for FRP-concrete interfacial bond strength prediction

Abstract

Globally, the technique of reinforcing concrete structures with bonded fiber-reinforced polymers (FRP) has become widely adopted. The integrity of the interface between concrete and FRP significantly influences the behavior of the reinforced structure. Consequently, precise prediction of the bond strength at the concrete and FRP interface is crucial for the logical design and assessment of structures that are repaired and reinforced using FRP. This paper utilizes two emerging metaheuristic algorithms, the Slime Mould Algorithm (SMA) and the Dung Beetle Optimization Algorithm (DBO), to improve the performance of machine learning (ML) techniques, including KNN, SVR, GBDT, and XGBoost. Optimizing the ML models with metaheuristic algorithms significantly enhanced the prediction accuracy compared to the non-optimized models. The SMA-GBDT performed better than other ML models, achieving an R² of 0.9492, an MAE of 1.5294, an MSE of 6.4159, an RMSE of 2.5329, and a MAPE of 8.6916, based on the testing dataset. Specifically, the SMA-GBDT model exhibited improvements of 5.83%, 39.04%, 50.75%, 29.82%, and 43.84% in R², MAE, MSE, RMSE, and MAPE, respectively, compared to the non-optimized GBDT. The predictions made by the SMA-GBDT model were higher precision than those provided by the current design codes and existing models.