A benchmark comparison and optimization of Gaussian process regression, support vector machines, and M5P tree model in approximation of the lateral confinement coefficient for CFRP-wrapped rectangular/square RC columns

Abstract

In this study, various soft-computing models (Gaussian process regression (GPR) and support vector machines (SVM) based on the polynomial kernel function (PKF), Pearson VII universal kernel function (PUKF), and radial basis kernel function (RBKF), as well as pruned/unpruned M5P tree models) were simultaneously applied for the first time in prediction of the lateral confinement coefficient (Ks) of CFRP-wrapped rectangular/square (R/S) RC columns, and their corresponding predictive successes were appraised statistically. For this aim, short side of the column section (b), long side of the column section (h), total thickness of CFRP (t), compressive strength of the unconfined concrete (f'c0), and the elastic modulus of CFRP (ECFRP) were used as independent input variables whereas the Ks was the output variable. Results indicated that the performance of the Pearson VII kernel function-based Gaussian process regression (GPR-PUKF) model was superior to other models for the training and testing stages. A sensitivity investigation showed that the total thickness of CFRP (t) was the most effective parameter for predicting the Ks using GPR-PUKF-based model. Findings of the present computational assessment obviously revealed that the employed soft-computing strategy had the capability of accurately estimating the Ks of R/S RC columns wrapped with CFRP.