Data mining methods for environmental applications

Signal processing, data analysis and pattern recognition oriented to environmental problems.

Using a large set of modern data mining methods, TTSS team is able to derive significant information from acquired time-series which can be used as feature descriptors of underlying processes. We focus on the extraction of information patterns that can be used as building elements of machine learning models which allow predictive estimations of future behaviors. Representative methods that we used, are:

  • 1D and 2D wavelet Transforms
  • Natural time analysis
  • Graph networks representations
  • Entropy based information measures of complex systems
  • Classification trees
  • Support vector machines
  • Ensemble methods

Example applications include:

  • Earthquake Early Warning Systems
  • Estimation of critical points in SOC (Self organized criticality) systems
  • Study of dynamic characteristics at fractal and multifractal time series
  • Complex systems with long-range interactions and metastable states
  • Development of soft-sensors for non-destructive assesment
  • Nonlinear signal processing
  • Denoising methods
  • Compression and Dimensionality reduction representations