Enhancing%20Interactive%20Visual%20Data%20Analysis%20by%20Statistical%20Functionality

1
Enhancing Interactive Visual Data Analysis by
Statistical Functionality

• Jürgen Platzer
• VRVis Research Center
• Vienna, Austria

2
Overview

• Motivation
• Statistics Library for Information Visualization
• Sample Application
• Conclusions

3
Motivation

• Information visualization and statistical methods
  try to enable a better insight into data
• The same goal is reached by different means

• Statistical Routines

Information Visualization

• Users pattern recognition system
• Creates interactively modifiable graphics
• Allows interactive efficient information
  drill-down
• Low dimensional features are easily detected and
  analyzed.
• Linked views allow interactive investigation of
  functional coherences.

• Todays computational possibilities
• Computation of facts, summaries, models, ...
• A large variety of algorithms for specific tasks
  (clustering, dimension reduction,...)
• Based on the knowledgeable theory of data
  exploration
• Considers multivariate relationships
• Results can be easily reproduced

4
Aim of this work

• Put users input and algorithmic capabilities on
  the same level.
• Let them interactively communicate
• Show that the interactive combination of the
  strength of both fields makes visual data mining
  more efficient.

5
Statistics Library for InfoViz

• Find the most important statistical functions for
  explorative data analysis.
• Clustering (Hierarchical approaches, partitional
  heuristics)
• Dimension reduction (MDS, PCA, SOM)
• Transformation of Dimensions (Linear vs.
  non-linear)
• Statistical Moments (classic vs. robust)
• Regression

6
Statistics Library for InfoViz

• Additionally include innovative concepts
• Robustness
• Reduce influence of outliers
• Detect outliers
• Integration of multivariate outlier
  identification
• Fuzzyness
• Data comes from real world
• The real world is not based on bits!-)
• Integrate uncertainty in clustering by fuzzy k
  means

7
Statistics Library for InfoViz

• Fuzzy k means (UVW dataset - 149 769 data items)

8
Statistics Library for InfoViz

• Create hooks of interaction
• Allow the interactive communication between
  algorithm and the user.
• Immediate updates of summaries based on
  selections
• Translation of user action into parameter
  settings
• Starting algorithms based on previous results

9
Sample Application

• Interactive Clustering (Letter image recognition
  data 4640 data items, 6 groups)

10
Sample Application

• Interactive Clustering (Letter image recognition
  data 4640 data items, 6 groups)

11
Sample Application

• Interactive Clustering (Letter image recognition
  data 4640 data items, 6 groups)

12
Conclusions

• Keyword INTERACTIVITY
• Immediate validation of results
• Immediate adaptation of algorithms
• Immediate numerical feedback of user actions
• Information exchange user / algorithm
  incorporation of multivariate features
• Research of possible communication concepts
  between user and statistical algorithms
• Translation of user actions into parameter
  settings

13
Acknowledgement

• Peter Filzmoser
• Helwig Hauser
• Harald Piringer
• Austrian research program Kplus

14
Thank you for your attention.Are there any
questions?