ROBUST MAPPING OF SPATIAL DATA WITH SUPPORT VECTOR REGRESSION (PREPRINT IBRAE-2002-18)

Preprint IBRAE-2002-18

Pozdnukhov A., Kanevski M., Maignan M., Canu S.

The paper is devoted to the description of the Support Vector Regression - a model based on the Statistical Learning Theory. The methodology of application of the SVR models to the problem of spatial data prediction mapping is considered on a real case study: soil contamination with ¹³⁷Cs radionuclide in Briansk region. The procedure of SVR hyper-parameters tuning is considered in details. The cases of noisy data and data with outliers were considered. A way of incorporation the additional information on measurements quality is discussed. Preliminary results on multi-scale SVR modeling are presented.

Bibliographical reference 

Pozdnukhov A., Kanevski M., Maignan M., Canu S. ROBUST MAPPING OF SPATIAL DATA WITH SUPPORT VECTOR REGRESSION. Preprint IBRAE-2002-18. Moscow: Nuclear Safety Institute RAS, July 2002. 17 p. — Refs.: 14 items.