Explainable machine learning model and gene expression programming for predicting reinforced concrete beams moment capacity exposed to fire.

Abstract

In this study, a new formulation for the moment capacity (M_r) of Reinforced Concrete (RC) beams under fire conditions is estimated using Gene Expression Programming (GEP). In addition, the use of Machine Learning (ML) methods such as XGBoost, AdaBoost, and LightGBM is investigated for estimating the M_r of RC beams in fire. The database for predicting the M_r of RC beams includes 280 samples. In this paper, the cross-section width b_w, cross-section depth d, distance from the beam edge to the center of steel reinforcement d_eff, area of steel reinforcement A_st, time duration of fire t, compressive strength of concrete f_c, and moment capacity of the beam under fire M_r are considered as the parameters of ML models. Several statistical metrics were employed to assess the performance of the models, including the mean absolute error (MAE), mean square error (MSE), root mean square error (RMSE), coefficient of determination (R²), and gradients of regression lines (k and k'). In this study, Shapley Additive exPlanations (SHAP) analysis was used to interpret the predictions of the XGBoost model, which was selected for its high accuracy with the best R² and the lowest error rate. The results indicate that the methods used demonstrate high accuracy in estimating the M_r of RC beams.