Application of Tree-Based Machine Learning Methods in Predicting the Shear Capacity of Steel Reinforced Concrete Beams without Stirrups

Abstract

This study proposes two tree-based machine learning models, namely Ensemble Learning (ELB) and Random Forest (RF), to predict the shear resistance of reinforced concrete beams without reinforcement. A database of 1849 beam test results collected from the available literature is used for the training and validation phases of the proposed tree models. The database uses twelve input parameters, representing the beam’s geometry, loading conditions, and material properties. The evaluation of the models is performed using the cross-validation technique and well known statistical criteria, namely the coefficient of determination (R2), root mean square error (RMSE), and mean absolute error (MAE). The results show that both models can perform well in predicting the shear resistance of reinforced concrete beams without reinforcement, with R2 = 0.917, RMSE = 43.32, MAE = 20.82 using ELB model, and R2 = 0.913, RMSE = 46.4, MAE = 22.43 for RF model. These excellent results demonstrate that the proposed tree-based machine learning models are accurate and useful predictors for engineers in the pre-design phase.