Visual Data Mining: An Overview

What is Visual Data Mining? Survey of techniques

Data Visualization Visualizing Data Mining Results Visual Data Mining

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

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

– +

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

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

Computer Graphics

High Performance Computing

Pattern Recognition

Human Computer Interfaces

Multimedia Systems

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

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

abilities of

the computer

General KnowledgeCreativity

Logic

Data Storage

Numerical Computation

Planning

PredictionDiagnosis

Searching

Perception

human abilities

Abilities of Humans and Computers

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

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.

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

Geometric

Icon-based

Pixel-oriented

Hierarchical

Graph-based

Mapping Projection Filtering Link & Brush Zooming

Simple

Complex

Data Visualization Techniques

Distortion Techniques

Interaction Techniques

Dimensions of Exploratory Data Visualization

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

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,...)

Visual Survey

Data visualization techniques Scatterplot Matrices, Landscapes, Parallel

Coordinates Icon-based, Dimensional Stacking, Treemaps

Direct Visualization

Ribbons w

ith Tw

ists Based on V

orticity

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]

matrix of scatterplots (x-y-diagrams) of the k-dimensional data [total of (k2/2-k) scatterplots]

Use

d by

erm

issi

on o

f M

. W

ard,

Wor

cest

er P

olyt

echn

ic In

stitu

te

Scatterplot-Matrices [Cleveland 93]

Landscapes [Wis 95]

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

news articlesvisualized asa landscape

Use

d b

y p

erm

issi

on

of

B. W

rig

ht,

Vis

ible

Dec

isio

ns

Inc.

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

Attr. 1 Attr. 2 Attr. kAttr. 3

• • •

Parallel Coordinates

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)

census data showing age, income, sex, education, etc.

used

by

perm

issi

on o

f G

. G

rinst

ein,

Uni

vers

ity o

f M

assa

chus

ette

s at

Low

ell

Stick Figures

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]

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

attribute 1

attribute 2

attribute 3

attribute 4

Used by permission of M. Ward, Worcester Polytechnic InstituteVisualization 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

Dimensional Stacking

Dimensional Stacking

Disadvantages: Difficult to display more than nine

dimensions Important to map dimensions

appropriately May be difficult to understand

visualizations at first

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)

Treemap [JS 91, Shn 92, Joh 93]

MSR Netscan image:

Treemap of a File System (Schneiderman)

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)

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

Boxplots from Statsoft: Multiple Variable Combinations

Visualization of Data Mining Results in SAS Enterprise Miner: Scatter

Plots

Visualization of Association Rules in SGI/MineSet 3.0

Visualization of Decision Tree in SGI/MineSet 3.0

Vizualization of Decision Trees

Visualization of Cluster Grouping IBM Intelligent Miner

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

MineSet: Association Rules

Association Ball Graph (DBMiner)

Items are visualized as balls

Arrows indicate rule implication

Size represents support

Classification (SAS EM [SAS 01])

Color corresponds to relative frequency of a class in a node

Branch line thickness is proportional to the square root of the objects

Tree Viewer

Cluster Analysis (H-BLOB: Hierarchical BLOB) [SBG 00]

Cluster Form ellipsoids Form blobs(implicit surfaces)

H-BLOB

Text Mining (ThemeRiver [WCF+ 00])

Visualization of thematic Changes in documents Vertical distance indicates collective strength of the themes

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

Visualization of Data Mining Processes by Clementine

Understand variations with visualized data

See your solution discovery process clearly

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

Visual data mining

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

’99]

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

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

Class-Preserving Projection:Preserves distances between projected means

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”

Interactive Visual Mining by Perception-Based Classification

(PBC)

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

Visual Classification

Visual Classification Results

Comparable classification accuracy Can produce more understandable decision

trees Expert domain knowledge can be exploited

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

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

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)