Improving Prediction of Compressive Strength of Rectangular/Square (R/S) FRP-Confined Concrete Using Machine Learning

Several experimental studies have been conducted on concrete confined with FRP sheets, and various models have been proposed in previous research to determine its compressive strength. However, studies have shown that Machine Learning (ML)-based methods offer higher accuracy than these models. In this study, the effectiveness of different machine learning methods is investigated for predicting the ultimate compressive strength of Rectangular/Square (R/S) FRP-confined concrete columns. These methods include ELM, GMDH, ANFIS, and the Kriging interpolation method. Also, this study proposes utilizing optimization science as a solution to enhance the performance of the ANFIS method. As an innovation in this study, the Marine Predators Algorithm (MPA), a nature-inspired metaheuristic, has been used to optimize the parameters of the ANFIS method. To show the ability of ML methods to estimate compressive strength, statistical indices were calculated and ML methods were compared; the correlation coefficient (R۲) for ELM, GMDH, ANFIS, ANFIS-MPA, and Kriging interpolation methods was equal to ۰.۸۹, ۰.۹۲, ۰.۹۲, ۰.۹۳, and ۰.۹۸, respectively. Also, these results show that the proposed methods have better performance than the best models in previous studies, with an average error reduction of ۶۲%.

کلیدواژه ها:

Extreme Learning Machine ، adaptive neural fuzzy inference system ، Marine Predators Algorithm ، Kriging ، Compressive Strength