1. Visual Data Mining: An Overview
What is Visual Data Mining?
Survey of techniques
Data Visualization
Visualizing Data Mining Results
Visual Data Mining

2. What Is Visual Data Mining?
Visual data mining “discovers implicit and useful knowledge from large data sets using data and/or knowledge visualization techniques”
Data visualization + Data mining techniques
From Han and Kamber “Data Mining” p. 463

3. Why Visual Data Mining? Advantages of human visual system
Highly parallel processor
Sophisticated reasoning engine
Large knowledge base
Can be used to comprehend data distributions, patterns, clusters, and outliers
Data Mining Algorithms
Visualization
Actionable + –
Evaluation
Flexibility
User Interaction

4. Why Not Only Visual Data Mining?
Disadvantages of human visual system
Needs training
Not automated
Intrinsic bias
Limit of about 106 or 107 observations (Wegman 1995)
Power of integration with analytical methods
Stat from

5. Scope of Visual Data Mining
Visualization: Use of computer graphics to create visual images which aid in the understanding of complex, often massive representations of data
Visual Data Mining: The process of discovering implicit but useful knowledge from large data sets using visualization techniques
Human Computer Interfaces
Computer Graphics
Multimedia Systems
High Performance Computing
Pattern Recognition

6. Purpose of Visualization
Gain insight into an information space by mapping data onto graphical primitives
Provide qualitative overview of large data sets
Search for patterns, trends, structure, irregularities, relationships among data
Help find interesting regions and suitable parameters for further quantitative analysis
Provide a visual proof of computer representations derived

7. Visual Data Mining & Data Visualization
Integration of visualization and data mining
data visualization
data mining result visualization
data mining process visualization
interactive visual data mining
Data visualization
Data in a database or data warehouse can be viewed
at different levels of abstraction
as different combinations of attributes or dimensions
Data can be presented in various visual forms

8. Abilities of Humans and Computers

9. Visual Mining vs. Scientific Vis. & Graphics
Scientific Visualization
Often visualize physical model, low dimensionality
Graphics
More concerned with how to render (draw) rather than what to render

10. Data Visualization View data in database or data warehouse
User may control
Different levels of details
Subset of attributes
Drawn using boxplots, histograms, polylines, etc.

11. Historical Overview of Exploratory Data Visualization Techniques (cf
Historical Overview of Exploratory Data Visualization Techniques (cf. [WB 95])
Pioneering works of Tufte [Tuf 83, Tuf 90] and Bertin [Ber 81] focus on
Visualization of data with inherent 2D-/3D-semantics
General rules for layout, color composition, attribute mapping, etc.
Development of visualization techniques for different types of data with an underlying physical model
Geographic data, CAD data, flow data, image data, voxel data, etc.
Development of visualization techniques for arbitrary multidimensional data (w.o. an underlying physical model)
Applicable to databases and other information resources

12. Dimensions of Exploratory Data Visualization

13. Classification of Data Visualization Techniques
Geometric Techniques:
Scatterplots, Landscapes, Projection Pursuit, Prosection Views, Hyperslice, ParallelCoordinates...
Icon-based Techniques:
Chernoff Faces, Stick Figures, Shape-Coding, Color Icons, TileBars,...
Pixel-oriented Techniques:
Recursive Pattern Technique, Circle Segments Technique, Spiral- & Axes-Techniques,...
Hierarchical Techniques:
Dimensional Stacking, Worlds-within-Worlds,Treemap, Cone Trees, InfoCube,...
Graph-Based Techniques:
Basic Graphs (Straight-Line, Polyline, Curved-Line,...)
Specific Graphs (e.g., DAG, Symmetric, Cluster,...)
Systems (e.g., Tom Sawyer, Hy+, SeeNet, Narcissus,...)
Hybrid Techniques: arbitrary combinations from above

14. Distortion & Dynamic/Interaction Techniques
Distortion Techniques
Simple Distortion (e.g. Perspective Wall, Bifocal Lenses, TableLens, Graphical Fisheye Views,...)
Complex Distortion (e.g. Hyperbolic Repr. Hyperbox,...)
Dynamic/Interaction Techniques
Data-to-Visualization Mapping (e.g. Auto Visual, S Plus, XGobi, IVEE,...)
Projections: (e.g. GrandTour, S Plus, XGobi,...)
Filtering (Selection, Querying) (e.g. MagicLens, Filter/Flow Queries, InfoCrystal,...)
Linking & Brushing (e.g. Xmdv-Tool, XGobi, DataDesk,...)
Zooming (e.g. PAD++, IVEE, DataSpace,...)
Detail on Demand (e.g. IVEE, TableLens, MagicLens, VisDB,...)

15. Visual Survey Data visualization techniques
Scatterplot Matrices, Landscapes, Parallel Coordinates
Icon-based, Dimensional Stacking, Treemaps

16. Direct Visualization Ribbons with Twists Based on Vorticity

17. Geometric Techniques Basic Idea
Visualization of geometric transformations and projections of the data
Methods
Landscapes [Wis 95]
Projection Pursuit Techniques [Hub 85] (a techniques for finding meaningful projections of multidimensional data)
Scatterplot-Matrices [And 72, Cle 93]
Prosection Views [FB 94, STDS 95]
Hyperslice [WL 93]
Parallel Coordinates [Ins 85, ID 90]

18. Scatterplot-Matrices [Cleveland 93]
Used by ermission of M. Ward, Worcester Polytechnic Institute
matrix of scatterplots (x-y-diagrams) of the k-dimensional data [total of (k2/2-k) scatterplots]

19. Landscapes [Wis 95] Visualization of the data as perspective landscape
news articles visualized as a landscape
Used by permission of B. Wright, Visible Decisions Inc.
Visualization of the data as perspective landscape
The data needs to be transformed into a (possibly artificial) 2D spatial representation which preserves the characteristics of the data

20. Parallel Coordinates [Ins 85, ID 90]
n equidistant axes which are parallel to one of the screen axes and correspond to the attributes
the axes are scaled to the [minimum, maximum]―range of the corresponding attribute
every data item corresponds to a polygonal line which intersects each of the axes at the point which corresponds to the value for the attribute

21. Parallel Coordinates

22. Icon-Based Techniques
Basic Idea
Visualization of the data values as features of icons
Overview
Chernoff-Faces [Che 73, Tuf 83]
Stick Figures [Pic 70, PG 88]
Shape Coding [Bed 90]
Color Icons [Lev 91, KK 94]
TileBars [Hea 95] (use of small icons representing the relevance feature vectors in document retrieval)

23. Stick Figures census data showing age, income, sex, education, etc.
used by permission of G. Grinstein, University of Massachusettes at Lowell
“two attributes mapped to axes, remaining attributes mapped to angle or length of limbs”. Look at texture pattern

24. Hierarchical Techniques
Basic Idea:  Visualization of the data using a hierarchical partitioning into subspaces.
Overview
Dimensional Stacking [LWW 90]
Worlds-within-Worlds [FB 90a/b]
Treemap [Shn 92, Joh 93]
Cone Trees [RMC 91]
InfoCube [RG 93]

25. Dimensional Stacking [LWW 90]
partitioning of the n-dimensional attribute space in 2-dimensional subspaces which are ‘stacked’ into each other
partitioning of the attribute value ranges into classes the important attributes should be used on the outer levels
adequate especially for data with ordinal attributes of low cardinality

26. Dimensional Stacking
Used by permission of M. Ward, Worcester Polytechnic Institute
Visualization of oil mining data with longitude and latitude mapped to the outer x-, y-axes and ore grade and depth mapped to the inner x-, y-axes

27. Dimensional Stacking Disadvantages:
Difficult to display more than nine dimensions
Important to map dimensions appropriately
May be difficult to understand visualizations at first

28. Treemap [JS 91, Shn 92, Joh 93]
Screen-filling method which uses a hierarchical partitioning of the screen into regions depending on the attribute values
The x- and y-dimension of the screen are partitioned alternately according to the attribute values (classes)
Picture:
MSR Netscan image:

30. Treemap of a File System (Schneiderman)

31. Treemaps
The attributes used for the partitioning and their ordering are user-defined (the most important attributes should be used first)
The color of the regions may correspond to an additional attribute
Suitable to get an overview over large amounts of hierarchical data (e.g., file system) and for data with multiple ordinal attributes (e.g., census data)

32. Data Mining Result Visualization
Presentation of the results or knowledge obtained from data mining in visual forms
Examples
Scatter plots and boxplots (obtained from descriptive data mining)
Decision trees
Association rules
Clusters
Outliers
Generalized rules
Text mining

33. Boxplots from Statsoft: Multiple Variable Combinations

34. Visualization of Data Mining Results in SAS Enterprise Miner: Scatter Plots

35. Visualization of Association Rules in SGI/MineSet 3.0

36. Visualization of Decision Tree in SGI/MineSet 3.0

37. Vizualization of Decision Trees
Picture of MineSet decision tree visualizer

38. Visualization of Cluster Grouping IBM Intelligent Miner

39. Association Rules (MineSet)
LHS and RHS items are mapped to x-, y-axis
Confidence, support correspond to height of the bar or disc, respectively
Interestingness is mapped to Color

40. MineSet: Association Rules

41. Association Ball Graph (DBMiner)
Items are visualized as balls
Arrows indicate rule implication
Size represents support

42. Classification (SAS EM [SAS 01])
Tree Viewer
Color corresponds to relative frequency of a class in a node
Branch line thickness is proportional to the square root of the objects

43. Cluster Analysis (H-BLOB: Hierarchical BLOB) [SBG 00]
Form blobs
(implicit surfaces)
Form ellipsoids

44. H-BLOB

45. Text Mining (ThemeRiver [WCF+ 00])
Visualization of thematic Changes in documents
Vertical distance indicates collective strength of the themes

46. Data Mining Process Visualization
Presentation of the various processes of data mining in visual forms so that users can see the flow of data cleaning, integration, preprocessing, mining
Data extraction process
Where the data is extracted
How the data is cleaned, integrated, preprocessed, and mined
Method selected for data mining
Where the results are stored
How they may be viewed

47. Visualization of Data Mining Processes by Clementine
See your solution discovery process clearly
Understand variations with visualized data
How the interactive Clementine knowledge discovery process works
See your solution discovery process clearly The interactive stream approach to data mining is the key to Clementine's power. Using icons that represent steps in the data mining process, you mine your data by building a stream - a visual map of the process your data flows through. Start by simply dragging a source icon from the object palette onto the Clementine desktop to access your data flow. Then, explore your data visually with graphs. Apply several types of algorithms to build your model by simply placing the appropriate icons onto the desktop to form a stream.
Discover knowledge interactively Data mining with Clementine is a "discovery-driven" process. Work toward a solution by applying your business expertise to select the next step in your stream, based on the discoveries made in the previous step. You can continually adapt or extend initial streams as you work through the solution to your business problem.
Easily build and test models All of Clementine's advanced techniques work together to quickly give you the best answer to your business problems. You can build and test numerous models to immediately see which model produces the best result. Or you can even combine models by using the results of one model as input into another model. These "meta-models" consider the initial model's decisions and can improve results substantially.
Understand variations in your business with visualized data Powerful data visualization techniques help you understand key relationships in your data and guide the way to the best results. Spot characteristics and patterns at a glance with Clementine's interactive graphs. Then "query by mouse" to explore these patterns by selecting subsets of data or deriving new variables on the fly from discoveries made within the graph.
How Clementine scales to the size of the challenge The Clementine approach to scaling is unique in the way it aims to scale the complete data mining process to the size of large, challenging datasets. Clementine executes common operations used throughout the data mining process in the database through SQL queries. This process leverages the power of the database for faster processing, enabling you to get better results with large datasets.

48. Interactive Visual Data Mining
Using visualization tools in the data mining process to help users make smart data mining decisions
Example
Display the data distribution in a set of attributes using colored sectors or columns (depending on whether the whole space is represented by either a circle or a set of columns)
Use the display to which sector should first be selected for classification and where a good split point for this sector may be

49. Visual data mining Projection Pursuits
(Class) Tours [Dhillon et al. ’98]
Visual Classification [Ankerst et al. KDD ’99]

50. Projection Pursuits Exploratory projection pursuit:
Goal: reduce dimensionality
Define “interestingness” index to each possible projection of a data set
Maximize this index, project linearly
Not always possible/useful

51. Class Tours
“Visualizing Class Structure of Multidimensional Data” by Dhillon et al. 1998
Problem: Visualize multidimensional data categorized into classes
Solution: Project data into 2D while preserving distances between class means

52. Class-Preserving Projection:
Preserves distances between projected means

53. Tours
Tours are animated and interpolated sequences of 2D projections [Asimov 1985]
Class tours: sequences of class-preserving 2-dimensional projections
Captures “inter-class structure of complex, multi-dimensional data”

54. Interactive Visual Mining by Perception-Based Classification (PBC)

55. Visual Classification
“Visual Classification: An Interactive Approach to Decision Tree Construction” by Ankerst et al. KDD 99
Exploit expert’s domain knowledge and human visual processing
Figure from

56. Visual Classification

57. Visual Classification Results
Comparable classification accuracy
Can produce more understandable decision trees
Expert domain knowledge can be exploited

58. Audio Data Mining
Uses audio signals to indicate the patterns of data or the features of data mining results
An interesting alternative to visual mining
An inverse task of mining audio (such as music) databases which is to find patterns from audio data
Visual data mining may disclose interesting patterns using graphical displays, but requires users to concentrate on watching patterns
Instead, transform patterns into sound and music and listen to pitches, rhythms, tune, and melody in order to identify anything interesting or unusual

59. Summary Many visualization methods available
How to evaluate and compare methods?
Need for:
Integrated visualization/exploration systems
Studies of interaction techniques for mining
Practical case studies

60. Acknowledgments
Many slides and images from Mihael Ankerst, Boeing, Daniel A. Keim, AT&T, Tutorial at PKDD'2001
Some pictures from Information Visualization in Data Mining and Knowledge Discovery, edited by Usama Fayyad, Georges Grinstein and Andreas Wierse
A good set of slides were prepared by Andrew Wu (Spring 2004)