Evaluation of PredictivePotential of a learning Machine Method on Peak Outflow Due to Embankment Dam Failure

In this study, the potential of Support Vector Machine (SVM (in the prediction of peak outflow due to dam failure was evaluated. Huge volume deposited behind dam may cause a large casualty in case of sudden release of it to downstream. Head and volume of water at the time of failure were considered as inputs to SVM model. To train these models, 70% of 112 gathered data from literature was used as training subset and the rest 30% was used to test the model as test subset, at the same time these two subsets were selected in a way to be statistically similar. After studying four SVM models with different kernel functions, i.e. Linear, Polynomial, Radial Basis Function and Sigmoid, it was found that SVM with Radial Basis Kernel function outperforms other models. Results of the statistical evaluation of the proposed model are satisfying with R2=0.96, RMSE=0.03 for training phase and R2=0.94, RMSE=0.05 for the test phase. A comparison was made between some conventional empirical equations and the proposed model in which results shows proposed SVM model surpasses empirical equations in predicting peak outflow due to embankment dam failure.