Development of Machine Learning-based Methods to Reduce the Uncertainty of Tunneling Projects

Abstract

Having a good knowledge of the time and cost required to build a tunnel can be very important in reducing uncertainties related to the management of its construction. In this paper, using data obtained from the constructed parts of a tunnel, Gaussian process regression (GPR) method is developed to predict the time and cost of the non-constructed parts. Finally, by comparing the results predicted by the GPR model with the actual ones, it was concluded that the developed GPR model has a high potential to reduce uncertainties related to the time and cost of tunnel construction. Also, the ability of GPR model to predict time and cost of tunnel construction was compared with two other methods of support vector regression (SVR) and artificial neural networks (ANN). Finally, the GPR model was superior to the SVR and ANN methods in terms of prediction accuracy