SVRT: a decision tree-assisted support vector regression for modeling engineering data with complex regression relationship

Support vector regression has been widely used in engineering data modeling. The regression relationship of engineering data usually varies greatly, which results that it being difficult to evaluate through a unified prediction model. In this work, a decision tree-assisted support vector regression is proposed, which can take advantage of training samples partition to improve prediction accuracy. A decision tree is proposed to partition the training samples in such a way that the samples in the same part have a more similar regression relationship to each other than to those in the other parts. Support vector regression is used to evaluate the regression relationship, and an optimization algorithm is designed to search the best splitting input variable and division point at each node. For a new sample to predict, it is compared from the root node of the decision tree until reaches a certain leaf node, and the response is obtained according to the prediction model contained in the leaf node. Experiments show that the proposed method is able to provide competitive prediction results compared with the conventional prediction methods.

Automated summary

This study proposes a decision tree-assisted support vector regression method, which can improve prediction accuracy by using training samples partition. Experimental results show that it has competitive prediction results compared with traditional methods.