Data-driven prediction of axial compression capacity of GFRP-reinforced concrete column using soft computing methods

In corrosive environments and the presence of chemical materials, steel reinforcements inside concrete columns are corroded. To maintain durability and prevent their corrosion, reinforcements with FRP polymer materials can be used. In this article, the axial behavior of concrete columns with Glass Fiber Reinforced Polymer (GFRP) reinforcements is investigated using Machine Learning (ML) models and analytical equations. The nine analytical equations proposed by researchers and analyzed based on laboratory results are considered in this study. Also, four ML models, including Artificial Neural Network (ANN), Gaussian Processes Regression (GPR), Support Vector Machine (SVM), and Multivariate Adaptive Regression Spline (MARS) were investigated to evaluate the experimental data. The value of R2 for ANN, GPR, SVM, and MARS models is 0.9940, 0.9897, 0.9869, and 0.9794, respectively. Root mean square error (RMSE), Mean absolute error (MAE), Mean absolute percentage error (MAPE), Nash-Sutcliffe model efficiency (NSE) and scatter index (SI) parameters were also investigated to evaluate the models. Based on these parameters, the ANN model provides the best accuracy and efficiency in predicting the axial capacity (AC) of GFRP reinforced concrete (RC) columns. On the other hand, the MARS model, with two fewer input variables than other models and without a significant loss of accuracy, allows for the prediction of AC using a simple model. Among the analytical equations, the model provided by Afifi et al. and Tobbi et al. show the best efficiency in the low errors.