Abstract
In the area of ground improvement, the stone columns (SCs) play a definite role. The ground treatment technique has been demonstrated to be effective in improving the embankments’ stability and natural slopes by raising the bearing capacity and decreasing settlements. The objectives of this study are to develop models for predicting the performance of SCs-supported embankment foundations utilizing artificial neural networks (ANN). For the aim of creating ANN models, training, testing, and validation sets comprising 70%, 15%, and 15% of the data, respectively steps were done, making use of available numerical results obtained from the 2D finite element analysis. A dataset including about 200 cases is involved, and the mean square error (MSE) with R-squared value is used as performance metrics of the system. The applied data in ANN models are arranged in the component of 4 input parameters, which cover column diameter d, centre-tocentre spacing S, the internal friction angle of columns material ϕ, and embankment high H. Relating to these input parameters, the selected responses were the bearing capacity of the SC (BC) and the safety factor against the stability (SF). Based on the simulated results, an ideal 4- 14-1 ANN architecture has been settled for the direct prediction. According to the technique used, the forecasted data from the model had a good agreement with the actual datum, where the high regression coefficient (R2) was equal to 0.995 and 0.891 for BC and SF models, respectively. Furthermore, the relative importance of influential variables is examined, which shows that the column diameter is the most effective parameter in the two study models with a significance score of 32.9%. Finally, the outcomes clearly demonstrated that the ANN method is reliable for modelling and optimizing of the SC behaviour.