The pollution of soils with heavy metals and radio-nuclides is a complex
phenomenon. The interdisciplinary approach is often required in modeling.
The measurements collected on soil pollution usually form a multivariate
dataset which includes information of different kinds. These are
the soil types, concentrations and radioactivity of pollutants which
comes in the form of both “hard” and “soft” measurements, the related
information on land cover type, etc. Given the growing amount of
such information, the algorithmic approaches to data modeling have
developed rapidly last years. In particular the methods based on
data mining and machine learning have been used in a growing number
of applications. These methods follow a data-driven methodology,
aiming at providing the best possible generalization and predictive
abilities for the given task. In this paper, the approach to multi-class
classification of soil types with machine learning methods such as
Probabilistic Neural Networks (PNN) and Support Vector Machines (SVM)
is presented. The task is important for the modeling of radio-nuclides
vertical migration, which properties are highly dependent on the
soil type. The real case study deals with the data on Chernobyl accident,
where high variability of environmental parameters and initial fallout
at different scales highly complicates the solution of the whole
problem of prediction mapping and risk assessment. Particularly,
the official soil type maps do not provide sufficient information
for modeling, while the information on soil types accompanies radioactivity
level measurements and can be used for soil types mapping. The PNN
and SVM methods are compared with the nearest neighbor method, the
simplest baseline approach to spatial classification.