Shear capacity estimation of reinforced concrete deep beam with vertical bar using artificial neural network and small datasets

In this study, we employ machine learning by applying an artificial neural network (ANN) to predict the shear capacity of simply supported reinforced concrete deep beams from a small dataset. A database of 76 experiments, comprising 13 key parameters, was prepared and used to train and tune various ANN configurations. The Levenberg−Marquardt algorithm converged fastest and most accurately after systematic trials and introducing a second hidden layer significantly enhanced the nonlinear mapping. An optimal network of 11-12 neurons with radial basis activation achieved a training root mean square error (RMSE) of 0.2345. Data validation revealed that correlation coefficients for training (0.999) and testing (0.992) were found, with over 95% of predictions within 5% of measured strengths. The model developed was shown to be overfitting as the number of datasets in this experiment is limited. Future studies need to be done to include more datasets to prevent overfitting.