Learning from and visualization of environmental, socio-economic and geodemographic data

The presentation gives an overview of recent studies carried out by the geomatics group of FGSE in the domain of environmental, natural hazards and socio-economic data analysis, modelling and visualization using machine learning algorithms (MLA). The main MLA recently used – artificial neural networks of different architectures (adaptive general regression neural networks, extreme learning machines, self-organizing maps) and a variety of kernel-based methods, have demonstrated their efficiency in studying multivariate and high-dimensional environmental phenomena. We present generic methodology and some of the results of real data case studies: environmental pollution, renewable resources assessment, natural hazards analysis (landslides, avalanches). Two particular topics of the current research will be discussed: 1) feature selection using extreme learning machines and simulated annealing; and 2) intrinsic dimension estimation and feature selection using the multipoint Morisita index.