An Approach Toward Numerical Data Augmentation and Regression Modeling Using Polynomial-Kernel-Based SVR

The paper is devoted to solving the actual problem of processing short datasets. The authors developed a new numerical data augmentation procedure that is based on the usage of polynomial-kernel-based SVR. The generalization property of the SVR method has increased during regression analysis of short datasets by using the following procedure. In addition, the paper presents the author's prediction procedure based on the extended dataset. This procedure provides a significant increase of the accuracy of the SVR with a polynomial kernel. The simulation of the method is based on the usage of a real short set of medical data. The optimal parameters of the method are selected; the efficiency of its work is compared with the existing methods of this class. The lowest errors of the developed method in comparison with the existing ones are established. Prospects for further research are presented in terms of improving the proposed approach and the new areas of its practical application.