Machine Learning Algorithm for Environmental Data Mining
The basic motivation of the current project is the explosive increase of environmental data of different nature: space-time monitoring, remote sensing images, in-situ measurements, etc., and, in parallel, recent fundamental developments in Machine Learning Algorithms (MLA).
Scientific problems can be formulated as follows:
- Development of a coherent and self-consistent methodology, based on MLA, for the recognition, modeling and prediction of structured patterns in environmental data.
- Construction and analysis of high dimensional input feature space by using expert knowledge and a variety of features selection/features extraction algorithms.
- Ensemble learning of data for better predictability.
- Exploration of dependencies in data.
- Development of brand-new feature selection techniques.
- Multi-task learning of multivariate data.
- Quantification of uncertainties for the intelligent decision-making process.
- Development of visual analytics tools for better understanding and communication of data and the results.
The applied part of the research deals with an application of the methodologies and the methods developed for simulated and modeled environmental risk and natural hazard data, e.g. pollution of the environment, landslides, wildfires, permafrost, and others.
Responsible: Mikhail Kanevski, Marj Tonini
Collaborators: Federico Amato, Fabian Guignard, Michael Leuenberger (Université de Neuchâtel), Carmen Vega Orozco (ex PhD student)
-
- Feature selection and modelling with Extreme Learning Machine
