Predicting the Maximum Load Capacity of Circular RC Columns Confined with Fibre-Reinforced Polymer (FRP) Using Machine Learning Model

Abstract

This article conducts an exhaustive investigation into the utilization of machine learning (ML) methods for forecasting the maximum load capacity (MLC) of circular reinforced concrete columns (CRCC) using Fiber-Reinforced Polymer (FRP). Extreme Gradient Boosting (XGB) algorithm is combined with novel metaheuristic algorithms, namely Sailfish Optimizer and Aquila Optimizer, to fine-tune its hyperparameters. The robustness and generalizability of these optimized hyperparameters are ensured through 200 Monte Carlo simulations (MCS). The model is constructed based on a database of 207 experimental results. Its performance is evaluated using three criteria: root mean squared error, mean absolute error, and the coefficient of determination. This study includes a performance comparison of the XGB4 model with eight other ML models, namely CatBoost (CAT), Gradient Boosting (GB), Hist Gradient Boosting (HGB), default XGB, Light Gradient Boosting (LGB), Linear Regression (LR), and Random Forest (RF). This comparison identifies the most effective model for predicting the MLC of columns. Additionally, this study explores the interpretability of the XGB model by SHAP values. This analysis illuminates the significance and interactions of various input features in predicting the FRP-confined CRCC's MLC. It offers insights into the primary elements influencing structural behavior by displaying a graphical depiction of the impact of specific characteristics on the model's output. This study culminates in developing an interactive Graphical User Interface (GUI) based on the XGB model. This tool allows users to investigate the influence of input parameters on the predicted MLC values, thereby enhancing their understanding and application of the model.