Abstract: To accurately predict the confinement effect of fiber-reinforced composite materials applied to concrete, strength and strain effect models for fiber-reinforced composite-constrained concrete were proposed based on Random Forest and CatBoost algorithms. Data of 502 concrete specimens with fiber-reinforced composite wrapping along the lateral side were collected, including parameters such as concrete cross-sectional dimensions, fiber-reinforced composite wrapping thickness, fracture strain and elastic modulus, unconstrained concrete strength and strain, constrained strength effect, and constrained strain effect. Existing empirical models for confinement strength and strain effects were compared and evaluated. Three evaluation metrics, namely determination coefficient (R²), root mean square error, and mean absolute error, were selected for the comparative evaluation of the developed Random Forest and CatBoost models. The results showed that for the confinement strength effect, the R² values of the Random Forest model and CatBoost model were generally above 0.84, higher than the existing empirical models (0.78 to 0.83). Regarding the confinement strain effect, the R² values of the Random Forest model and CatBoost model were generally above 0.87, also higher than the existing empirical models (0.65 to 0.75). This indicated that the performance of the developed Random Forest and CatBoost models was superior to the existing empirical models, and both models could provide relatively accurate predictions of the confinement effect of fiber-reinforced composite materials applied to concrete. Particularly, the CatBoost model demonstrated higher accuracy.

Key words: composite materials, fiber-reinforced, concrete, confinement strength effect, confinement strain effect