Frequency plot and relevance plot to enhance visual data exploration

  Frequency Plot and Relevance Plot to Enhance Visual Data

Exploration

José Fernando Rodrigues Jr.

Agma J. M. Traina

Caetano Traina Jr.

Computer Science Department

University of Sao Paulo - Brazil

http://www.icmc.usp.br/~junio/PublishedPapers/RodriguesJr_et_al_Frequency_Plot-SIBGRAPI2003.pdf

2/25

Outline

• The GBDIView Tool

• Frequency Plot with Interactive Filtering

• Relevance Plot

• Future Works and Conclusions

• Motivation• Motivation

• Visual Statistical Analysis

3/25

Motivation• Increasing volume of data that cannot be well utilized to

produce useful knowledge

• Raw Information Visualization techniques are limited in the task of data analysis

• Datasets might be unlimited both in size and complexity

• There is a need for visualization mechanisms that reduce the drawback of massive datasets.

The efficient use of the data can provide helpful insight in critical decision

making.

4/25

The Problem

• Massively populated datasets tend to result in a visualization scene with an unacceptable level of cluttering;

• Some regions of the data seam like blots in the visualization scene.

• Many Information Visualization techniques have already been proposed to attack these problems

• It is becoming each time more challengeable to create new ones.

5/25

Outline• Motivation

• The GBDIView Tool

• Frequency Plot with Interactive Filtering

• Relevance Plot

• Future Works and Conclusions

• The GBDIView Tool

• Visual Statistical Analysis

6/25

The GBDIView Tool

A preliminary version of a Visualization Environment, and a

partially working idea

7/25

The GBDIView Tool Features

• 4 well-known visualization techniques: Parallel Coordinates, Scatter Plots, Star Coordinates, and Table Lens

• Interaction with Link & Brush and interactive filtering

• Basic statistics presentation

• Enabled with Frequency Plot and Relevance Plot

8/25

Development

• Borland C++ Builder 5

• OpenGL

• Software Component

• Open source

Memory sharing and pipeline support.Highly reusable code.

9/25

Outline• Motivation

• The GBDIView Tool

• Frequency Plot with Interactive Filtering

• Relevance Plot

• Future Works and Conclusions

• Frequency Plot with Interactive Filtering

• Visual Statistical Analysis

10/25

Frequency Plot with Interactive Filtering

• A method that combines the selective filtering technique with an automatic statistical analysis

• The frequency here means how frequently a given attribute value can be found in a dataset

• The frequency is visually presented through the opacity of the graphical items

11/25

Example

12/25

The Breast Cancer Dataset(Cortesy by the University of California at Irvine Machine Learning Laboratory)

• 457 records

• 11 attributes: 1 sample identifier, 9 laboratorial results, 1 attribute for classification

•Attribute “CLASS”: 0 for benign cancer and 1 for malign

13/25

Comparison

• The Frequency Plot is comparatively more powerful than the raw visualization technique

• The probability analysis can reveal clusters in subsets of the dataset

• The behavior of the data is immediately characterized as the user interacts with it

14/25

Outline• Motivation

• The GBDIView Tool

• Frequency Plot with Interactive Filtering

• Relevance Plot

• Future Works and Conclusions

• Visual Statistical Analysis

• Relevance Plot

15/25

Relevance PlotThe data is presented accordingly to its relevance to a

user’s defined set of interesting points

X1

X1 = RP1 + MRD

Relevance = 0

X0 = RP0

Relevance = 1

X0

X2

X3

Null RP2 Not Considered

Dist = 1

Relevance = - 1

The relevance point is over the attribute value

The distance is equal the Maximum relevance

distance The distance is the maximum possible

Relevance = 1 + 0 + (-1) = 0/3 = 0

16/25

Example

17/25

• Provides an interactive fuzzy query in a visual environment

• Allows to discover items of interest in a speculative way

• Extends the interactive filtering approach

Features of the Relevance Plot

18/25

Outline• Motivation

• The GBDIView Tool

• Frequency Plot with Interactive Filtering

• Relevance Plot

• Future Works and Conclusions

• Visual Statistical Analysis• Visual Statistical Analysis

19/25

Visual Statistical Analysis

• Provides a summarization of the data being visualized

• Visually demonstrates meaningful features of the data

• Weaken the drawbacks of analysing too populated data sets

20/25

Visual Statistical Analysis

21/25

Outline• Motivation

• The GBDIView Tool

• Frequency Plot with Interactive Filtering

• Relevance Plot

• Future Works and Conclusions

• Visual Statistic Analysis

• Future Works and Conclusions

22/25

Future Work: Possibilities for Presentation

• Possibility of presentation through many visual effects as size, color hue and color brightness

• Color mappings and 3D effects (depth perception) might also be used

23/25

Future Work: Possibilities for Analysis

• Most basic schema: Euclidean distance, but other distance schemas might be used for additional insights

• Different distance calculus for different dimensions

• Weights for the dimensions

• Customization

24/25

Conclusions

• It is a challenge to discover new visualization techniques that, in raw format, can contribute to visual analysis

• Visualization techniques should be improved by automatic analysis mechanisms joined with interaction techniques

• The Frequency Plot and Relevance Plot methods can enhance visualization techniques of almost all kinds

25/25

The End

Thanks for coming