PSI - Issue 70

Rachit Sharma et al. / Procedia Structural Integrity 70 (2025) 386–393 Sharma and Laskar/ Structural Integrity Procedia 00 (2025) 000 – 000

390

5

Table 3. ML model hyperparameters. Number Model Parameter

Optimized value 4000/ 0.1/ rbf 0.09/ 1/ 900/ 3

1. 2.

SVR

C/ gamma/ kernel

XGBoost learning rate/ min_child_weight/ n_estimators/ max_depth

5. Result and Discussion 5.1. Prediction Performance of ML Models

The scatterplot for prediction performance of design codes ACI440.1R-15 and GB50608-2020 is shown in Fig. 2 (a) and (b). The plots have lower value of R 2 (less than 0.50) indicating significant conservative approaches adopted in design codes. The comparison of results obtained via ML model SVR and XGBoost is shown in Fig. 2(c) and (d). Both the developed ML models show R 2 value higher than 0.90 for training as well as testing dataset. Among ML models, the XGBoost has slightly higher R 2 (0.94) as compared to SVR model (0.93). Shear capacity predictions from ensemble models typically contained in ±20%, indicating a strong alignment for predicted and experimental values. It confirms a strong predictive performance of ensemble model for estimating shear capacities of FRP-RC beams.

Fig. 2. Comparison of Scatterplot for Predicted and Experimental Shear Capacities.

The model performance has been evaluated using multiple performance indices, including (R 2 ), root mean squared error (RMSE), mean absolute error (MAE), and mean absolute percentage error (MAPE), as defined in Equation (4):