Ensemble regression based on polynomial regression-based decision tree and its application in the in-situ data of tunnel boring machine

Regression is an important branch of engineering data mining tasks, aiming to establish a regression model to predict the output of interest based on the input variables. To meet the requirements of different missions, the engineering system usually changes its operation status so that the regression relationship between the output and input variables changes. In this paper, two ensemble regression methods are proposed based on polynomial regression and decision tree, in which sample space partition is used to improve the prediction accuracy and ensemble strategy is used to improve the performance robustness of the regression model. The first ensemble regression method (named PRB) is developed under the framework of bagging. The second ensemble regression method (named PRF) is similar to the first one, but feature randomness is introduced. At each node of the polynomial regression-based decision tree, the polynomial regression error is used to select the best splitting feature. The experiments on a series of mathematical functions and engineering datasets indicate that the proposed ensemble regression methods outperform the polynomial regression-based decision tree, the polynomial regression method, and the random forest method in most experiments. The proposed ensemble regression methods are applied to model the dataset of a tunnel boring machine, aiming to predict the earth pressure based on the operation parameters of the cutterhead. The results indicate that the proposed two ensemble regression methods produce more accurate prediction results, and the PRF method performs best in most experiments.