Shear strength prediction of reinforced concrete beams by baseline, ensemble, and hybrid machine learning models

Abstract

The shear strength of reinforced concrete (RC) beams is critical in the design of structural members. Developing an effective mathematical method for accurately estimating shear strength of RC beams is beneficial for civil engineers. This work presents a hybrid artificial intelligent (AI) model for effectively predicting the shear strength of various types of RC beam. The hybrid AI model was developed by integrating an optimization algorithm [smart firefly algorithm (SFA)] and machine learning [least squares support vector regression (LSSVR)], in which the SFA was used to optimize the hyperparameters of LSSVR, improving its predictive accuracy. Three large datasets were used to train and test the hybrid AI model in predicting shear strength of RC beams. The predictive accuracy of the hybrid AI model was compared comprehensively with those of single AI models, ensemble AI models, and empirical methods. The comparison results show that the hybrid AI model outperformed the others in predicting the shear strength of a wide range of RC beam types. In particular, with the test data of RC beams without stirrups, the hybrid AI model yielded a mean absolute percentage error (MAPE) of 21.703%. In predicting shear strength of RC beams with stirrups, the hybrid AI model yielded an MAPE of 12.941%. For RC beams with FRP reinforcement, the hybrid AI model yielded an MAPE 18.951%. Therefore, this hybrid AI model can be a better alternative method to help civil engineers in designing RC beams.