PSI - Issue 70

N. Vignesh Kumar et al. / Procedia Structural Integrity 70 (2025) 453–460

459

4.2 Sensitivity Analysis The sensitivity analysis across parametric models (LASSO, STEPWISE Regression) and non-parametric models (Regression Tree Ensemble, Gaussian Process Regression, and Support Vector Machine) exhibits the following parametric importance patterns for both CFRP and GFRP confinement systems. With respect to the model performance R2 shown in Fig 1., the sensitivity analysis across five statistical models (LASSO, Stepwise Regression, Regression Tree Ensemble, Gaussian Process Regression, and Support Vector Machine reveals the thickness of the FRP (t) and the unconfined concrete strength (fco) are the most influential parameters pertaining to the Gaussian Progress Regression model irrespective of the FRP type

4.3 GPR Uncertainty Quantification

a

b

Fig. 3. GPR Uncertainty Quantification Analysis for (a) CFRP; (b) GFRP Gaussian Process Regression offers a key advantage over deterministic models by quantifying prediction ambiguities, providing both point predictions and associated variances for confined concrete strength (fcc). This enables calculation of 95% prediction intervals (PIs), which indicate the probabilistic range for true values vistal for risk assessment in structural engineering. Visualizations from Fig. 3 shows that, for CFRP, PIs generally cover actual values but vary in width, highlighting higher uncertainty in sparser data regions. For GFRP, intervals are more consistent. Error histograms reveal GPR slightly overpredicts strength, with tighter errors for GFRP. These analyses help designers assess uncertainty and make risk informed decisions. 5 Conclusion Five machine learning models namely STEPWISE, LASSO, SVM, RTE and GPR were developed for predicting the compressive strength of FRP confined concrete specimens using the database from Realfonzo and Napoli (2011). Specimen geometry and material properties of both FRP and concrete are the input parameters for the above mentioned models. Specifically, the FRP tensile strength calculation by the use of strain efficiency factor ke = 0.31 recommended by Realfonzo and Napoli (2011) is the noteworthy input parameter among the other input parameters. The parametric models like LASSO and STEPWISE regression resulted a comprehendible equation for predicting the FRP confined concrete strength with a relatively lesser performance metrics when compared with non-parametric models.