Choosing representative data items: Kohonen, neural gas or mixture model?

  • Authors:

  • Anna Bartkowiak;Joanna Zdziarek;Niki Evelpidou;Andreas Vassilopoulos

  • Affiliations:

  • Institute of Computer Science, University of Wroclaw, Wroclaw, Poland;Institute of Computer Science, University of Wroclaw, Wroclaw, Poland;Remote Sensing Laboratory, Geology Dept., University of Athens, Athens, Greece;Remote Sensing Laboratory, Geology Dept., University of Athens, Athens, Greece

  • Venue:
  • Enhanced methods in computer security, biometric and artificial intelligence systems
  • Year:
  • 2005

Quantified Score

Hi-index 0.00

Visualization

Abstract

When analyzing the erosion risk of Kefallinia, Greece, we have faced the problem, how to choose representatives (prototypes) for a big data set. We consider 3 methods serving this purpose: 1 - Kohonen's self-organizing map (SOM), 2 - Neural gas (NG), and 3. Mixture model (MM) of Gaussian distributions. The representativeness of the derived prototype vectors is measured by the quantization error, as defined by Kohonen (1995). It appears that neural gas and mixture models surpass quite steadily the SOM method in providing better representatives. To obtain a more thorough insight into the results, we map the obtained prototype vectors onto planes obtained by the neuroscale mapping, which seems to be a convenient alternative to Sammon's mapping. The SOM codebook vectors are visualized in the same planes and linked by threads. This is shown for the Kefallinia erosion data from Greece.