EPBM Advance Rate Prediction Using Hybrid Feature Selection and Support Vector Regression Modeling

Abstract

Advance rate (AR) prediction is crucial for optimal mechanized tunneling performance. However, the type of input features used when developing AR prediction models vary greatly from study to study. In this paper, a hybrid automatic feature selection method is presented and demonstrated through the development of a support vector regression (SVR) model for AR prediction in Earth pressure balance machine (EPBM) tunnel construction. EPBM datasets are collected from a tunnel project in the city of Porto, Portugal. Irrelevant features whose values are constant for most of the time were first removed via constant and quasi-constant detection method (CQD). Sequential forward selection (SFS) was then performed to determine the best subset of features to develop the best performed model. The results showed that the SVR model successfully predicted AR using the selected features with squared correlation coefficient (R 2 ) of 0.919 and 0.884 for training and testing, respectively. The efficiency of the feature selection method is demonstrated by comparing the results of the SVR model with feature selection and the one without. It is proved that proposed method helps improve the accuracy of the predictions by 8% and 17% for training and testing, respectively.