Support Vector Regression for Automated Robust Spatial Mapping of Natural Radioactivity

Abstract

This paper presents an application of Support Vector Regression method for the prediction of an environmental variable such as the level of natural radioactivity. The basics of the method are described, and some practical considerations are presented, including the meaning of the methods parameters and their influence on the model. The use of the prior data is discussed. It is shown how to include the information on the variance of the measurements into the model. The use of cross-validation for tuning the parameters of the algorithm is presented. Some ideas for detecting the unusual training samples with SVR are discussed. Generally, the case study illustrates the usefulness of the considered approach for automated spatial mapping tasks in the presence of prior data.