Data Mining: Chapter 2: Getting to Know Your Data

Concepts and Techniques

 Data Objects and Attribute Types

— Chapter 2 —  Basic Statistical Descriptions of Data

 Data Visualization
Jiawei Han, Micheline Kamber, and Jian Pei
University of Illinois at Urbana-Champaign  Measuring Data Similarity and Dissimilarity
Simon Fraser University
 Summary
©2011 Han, Kamber, and Pei. All rights reserved.
1 2

Types of Data Sets Important Characteristics of Structured Data

 Record
 Relational records  Dimensionality
 Data matrix, e.g., numerical matrix,


crosstabs
Document data: text documents: term-
 Curse of dimensionality
frequency vector
 Transaction data  Sparsity
 Graph and network
 World Wide Web  Only presence counts
 Social or information networks
 Molecular Structures  Resolution
 Ordered TID Items

Patterns depend on the scale

 Video data: sequence of images 1 Bread, Coke, Milk
 Temporal data: time-series 
2 Beer, Bread
Sequential Data: transaction sequences
Distribution

3 Beer, Coke, Diaper, Milk 
 Genetic sequence data
 Spatial, image and multimedia: 4 Beer, Bread, Diaper, Milk
 Spatial data: maps 5 Coke, Diaper, Milk  Centrality and dispersion
 Image data:
 Video data:
3 4

Data Objects Attributes

 Data sets are made up of data objects.  Attribute (or dimensions, features, variables):
 A data object represents an entity. a data field, representing a characteristic or feature
 Examples: of a data object.
 E.g., customer _ID, name, address
 sales database: customers, store items, sales
 medical database: patients, treatments  Types:
 Nominal
 university database: students, professors, courses
 Binary
 Also called samples , examples, instances, data points,
objects, tuples.  Numeric: quantitative

 Data objects are described by attributes.  Interval-scaled

 Database rows -> data objects; columns ->attributes.  Ratio-scaled

5 6

1
 Attribute Types Numeric Attribute Types
 Nominal: categories, states, or “names of things”  Quantity (integer or real-valued)
 Hair_color = {auburn, black, blond, brown, grey, red, white}  Interval
marital status, occupation, ID numbers, zip codes
Measured on a scale of equal-sized units


 Binary
 Nominal attribute with only 2 states (0 and 1)  Values have order
 Symmetric binary: both outcomes equally important  E.g., temperature in C˚or F˚, calendar dates
 e.g., gender  No true zero-point
 Asymmetric binary: outcomes not equally important.  Ratio
e.g., medical test (positive vs. negative)
Inherent zero-point


 Convention: assign 1 to most important outcome (e.g., HIV
positive)  We can speak of values as being an order of
 Ordinal magnitude larger than the unit of measurement
 Values have a meaningful order (ranking) but magnitude between (10 K˚ is twice as high as 5 K˚).
successive values is not known.  e.g., temperature in Kelvin, length, counts,
 Size = {small, medium, large}, grades, army rankings monetary quantities
7 8

Discrete vs. Continuous Attributes Chapter 2: Getting to Know Your Data

 Discrete Attribute
 Has only a finite or countably infinite set of values
 Data Objects and Attribute Types
 E.g., zip codes, profession, or the set of words in a

collection of documents
 Sometimes, represented as integer variables  Basic Statistical Descriptions of Data
 Note: Binary attributes are a special case of discrete
attributes
 Data Visualization
 Continuous Attribute
 Has real numbers as attribute values

 E.g., temperature, height, or weight  Measuring Data Similarity and Dissimilarity

 Practically, real values can only be measured and
represented using a finite number of digits  Summary
 Continuous attributes are typically represented as
floating-point variables
9 10

Basic Statistical Descriptions of Data Measuring the Central Tendency

Mean (algebraic measure) (sample vs. population): 1 n
x


 Motivation x  xi 
Note: n is sample size and N is population size. n i 1 N
 To better understand the data: central tendency, n
Weighted arithmetic mean:
variation and spread

wx i i
 Trimmed mean: chopping extreme values x  i 1
 Data dispersion characteristics n

 median, max, min, quantiles, outliers, variance, etc.

 Median: w
i 1
i

 Middle value if odd number of values, or average of

 Numerical dimensions correspond to sorted intervals the middle two values otherwise
 Data dispersion: analyzed with multiple granularities  Estimated by interpolation (for grouped data):
of precision n / 2  ( freq )l
 Boxplot or quantile analysis on sorted intervals
median  L1  ( ) width
 Mode freqmedian
 Dispersion analysis on computed measures  Value that occurs most frequently in the data
 Folding measures into numerical dimensions  Unimodal, bimodal, trimodal
 Boxplot or quantile analysis on the transformed cube  Empirical formula: mean  mode  3  (mean  median)
11 12

2
Symmetric vs. Skewed Data Measuring the Dispersion of Data
 Median, mean and mode of symmetric
 Quartiles, outliers and boxplots

symmetric, positively and  Quartiles: Q1 (25th percentile), Q3 (75th percentile)

Inter-quartile range: IQR = Q3 – Q1
negatively skewed data 

 Five number summary: min, Q1, median, Q3, max

 Boxplot: ends of the box are the quartiles; median is marked; add
whiskers, and plot outliers individually
 Outlier: usually, a value higher/lower than 1.5 x IQR
 Variance and standard deviation (sample: s, population: σ)
 Variance: (algebraic, scalable computation)
1 n 1 n 2 1 n 2
(xi  x)2  n 1[ xi  ( xi ) ]
n n
1 1
s  2   (x   )2  x 2
2 2
positively skewed negatively skewed
n 1 i1 i 1 n i1 N i 1
i
N i 1
i

 Standard deviation s (or σ) is the square root of variance s2 (or σ2)

February 4, 2016 Data Mining: Concepts and Techniques 13 14

Boxplot Analysis Visualization of Data Dispersion: 3-D Boxplots

 Five-number summary of a distribution

 Minimum, Q1, Median, Q3, Maximum
 Boxplot
 Data is represented with a box
 The ends of the box are at the first and third
quartiles, i.e., the height of the box is IQR
 The median is marked by a line within the
box
 Whiskers: two lines outside the box extended
to Minimum and Maximum
 Outliers: points beyond a specified outlier
threshold, plotted individually

15 February 4, 2016 Data Mining: Concepts and Techniques 16

Properties of Normal Distribution Curve Graphic Displays of Basic Statistical Descriptions

 The normal (distribution) curve  Boxplot: graphic display of five-number summary

 From μ–σ to μ+σ: contains about 68% of the  Histogram: x-axis are values, y-axis repres. frequencies
measurements (μ: mean, σ: standard deviation)
 From μ–2σ to μ+2σ: contains about 95% of it  Quantile plot: each value xi is paired with fi indicating
 From μ–3σ to μ+3σ: contains about 99.7% of it that approximately 100 fi % of data are  xi
 Quantile-quantile (q-q) plot: graphs the quantiles of
one univariant distribution against the corresponding
quantiles of another
 Scatter plot: each pair of values is a pair of coordinates
and plotted as points in the plane

17 18

3
 Histogram Analysis Histograms Often Tell More than Boxplots

 Histogram: Graph display of

tabulated frequencies, shown as 40  The two histograms
bars 35 shown in the left may
 It shows what proportion of cases 30 have the same boxplot
fall into each of several categories representation
25
Differs from a bar chart in that it is
The same values

20 
the area of the bar that denotes the
value, not the height as in bar 15 for: min, Q1,
charts, a crucial distinction when the 10 median, Q3, max
categories are not of uniform width  But they have rather
5
 The categories are usually specified different data
0
as non-overlapping intervals of 10000 30000 50000 70000 90000
distributions
some variable. The categories (bars)
must be adjacent

19 20

Quantile Plot Quantile-Quantile (Q-Q) Plot

Graphs the quantiles of one univariate distribution against the
Displays all of the data (allowing the user to assess both


corresponding quantiles of another
the overall behavior and unusual occurrences)  View: Is there is a shift in going from one distribution to another?
 Plots quantile information  Example shows unit price of items sold at Branch 1 vs. Branch 2 for
 For a data xi data sorted in increasing order, fi each quantile. Unit prices of items sold at Branch 1 tend to be lower
than those at Branch 2.
indicates that approximately 100 fi% of the data are
below or equal to the value xi

Data Mining: Concepts and Techniques 21 22

Scatter plot Positively and Negatively Correlated Data

 Provides a first look at bivariate data to see clusters of
points, outliers, etc
 Each pair of values is treated as a pair of coordinates and
plotted as points in the plane

 The left half fragment is positively

correlated
 The right half is negative correlated

23 24

4
 Uncorrelated Data Chapter 2: Getting to Know Your Data

 Data Objects and Attribute Types

 Basic Statistical Descriptions of Data

 Data Visualization

 Measuring Data Similarity and Dissimilarity

 Summary

25 26

Data Visualization Pixel-Oriented Visualization Techniques

 Why data visualization?  For a data set of m dimensions, create m windows on the screen, one
 Gain insight into an information space by mapping data onto graphical for each dimension
primitives  The m dimension values of a record are mapped to m pixels at the
 Provide qualitative overview of large data sets corresponding positions in the windows
 Search for patterns, trends, structure, irregularities, relationships among  The colors of the pixels reflect the corresponding values
data
 Help find interesting regions and suitable parameters for further
quantitative analysis
 Provide a visual proof of computer representations derived
 Categorization of visualization methods:
 Pixel-oriented visualization techniques
 Geometric projection visualization techniques
 Icon-based visualization techniques
 Hierarchical visualization techniques
 Visualizing complex data and relations (a) Income (b) Credit Limit (c) transaction volume (d) age
27 28

Laying Out Pixels in Circle Segments Geometric Projection Visualization Techniques

 To save space and show the connections among multiple dimensions,  Visualization of geometric transformations and projections
space filling is often done in a circle segment of the data
 Methods
 Direct visualization
 Scatterplot and scatterplot matrices
 Landscapes
 Projection pursuit technique: Help users find meaningful
projections of multidimensional data
 Prosection views
 Hyperslice
(a) Representing a data record Parallel coordinates
(b) Laying out pixels in circle segment 
in circle segment
29 30

5
 Direct Data Visualization Scatterplot Matrices
Ribbons with Twists Based on Vorticity

Used by ermission of M. Ward, Worcester Polytechnic Institute

Matrix of scatterplots (x-y-diagrams) of the k-dim. data [total of (k2/2-k) scatterplots]

Data Mining: Concepts and Techniques 31 32

Landscapes Parallel Coordinates

 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
Used by permission of B. Wright, Visible Decisions Inc.

corresponding attribute
news articles
visualized as
 Every data item corresponds to a polygonal line which intersects each
a landscape of the axes at the point which corresponds to the value for the
attribute

• • •

 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 Attr. 1 Attr. 2 Attr. 3 Attr. k
33 34

Parallel Coordinates of a Data Set Icon-Based Visualization Techniques

 Visualization of the data values as features of icons

 Typical visualization methods
 Chernoff Faces
 Stick Figures
 General techniques
 Shape coding: Use shape to represent certain
information encoding
 Color icons: Use color icons to encode more information
 Tile bars: Use small icons to represent the relevant
feature vectors in document retrieval

35 36

6
 Chernoff Faces Stick Figure
A census data
 A way to display variables on a two-dimensional surface, e.g., let x be figure showing
eyebrow slant, y be eye size, z be nose length, etc. age, income,
 The figure shows faces produced using 10 characteristics--head gender,
eccentricity, eye size, eye spacing, eye eccentricity, pupil size, education, etc.
eyebrow slant, nose size, mouth shape, mouth size, and mouth
opening): Each assigned one of 10 possible values, generated using
Mathematica (S. Dickson)

 REFERENCE: Gonick, L. and Smith, W. The

Cartoon Guide to Statistics. New York: A 5-piece stick
Harper Perennial, p. 212, 1993 figure (1 body
 Weisstein, Eric W. "Chernoff Face." From and 4 limbs w.
MathWorld--A Wolfram Web Resource. different
mathworld.wolfram.com/ChernoffFace.html angle/length)
37 Two attributes mapped to axes, remaining attributes mapped to angle or length of limbs”. Look at texture pattern 38

Hierarchical Visualization Techniques Dimensional Stacking

 Visualization of the data using a hierarchical attribute 4

partitioning into subspaces attribute 2

 Methods attribute 3

 Dimensional Stacking attribute 1

 Worlds-within-Worlds  Partitioning of the n-dimensional attribute space in 2-D

 Tree-Map subspaces, which are ‘stacked’ into each other
Partitioning of the attribute value ranges into classes. The
Cone Trees


important attributes should be used on the outer levels.
 InfoCube  Adequate for data with ordinal attributes of low cardinality
 But, difficult to display more than nine dimensions
 Important to map dimensions appropriately
39 40

Dimensional Stacking Worlds-within-Worlds

Used by permission of M. Ward, Worcester Polytechnic Institute
 Assign the function and two most important parameters to innermost
world
 Fix all other parameters at constant values - draw other (1 or 2 or 3
dimensional worlds choosing these as the axes)
 Software that uses this paradigm

 N–vision: Dynamic
interaction through data
glove and stereo
displays, including
rotation, scaling (inner)
and translation
(inner/outer)
 Auto Visual: Static
Visualization of oil mining data with longitude and latitude mapped to the interaction by means of
outer x-, y-axes and ore grade and depth mapped to the inner x-, y-axes queries
41 42

7
 Tree-Map Tree-Map of a File System (Schneiderman)

 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)

MSR Netscan Image

Ack.: http://www.cs.umd.edu/hcil/treemap-history/all102001.jpg 43 44

InfoCube Three-D Cone Trees

 A 3-D visualization technique where hierarchical  3D cone tree visualization technique works
well for up to a thousand nodes or so
information is displayed as nested semi-transparent
First build a 2D circle tree that arranges its
cubes


nodes in concentric circles centered on the

 The outermost cubes correspond to the top level root node
data, while the subnodes or the lower level data  Cannot avoid overlaps when projected to
2D
are represented as smaller cubes inside the
G. Robertson, J. Mackinlay, S. Card. “Cone
outermost cubes, and so on


Trees: Animated 3D Visualizations of

Hierarchical Information”, ACM SIGCHI'91
 Graph from Nadeau Software Consulting
website: Visualize a social network data set
that models the way an infection spreads
from one person to the next
Ack.: http://nadeausoftware.com/articles/visualization
45 46

Visualizing Complex Data and Relations Chapter 2: Getting to Know Your Data
 Visualizing non-numerical data: text and social networks
Tag cloud: visualizing user-generated tags
Data Objects and Attribute Types


 The importance of
tag is represented
by font size/color  Basic Statistical Descriptions of Data
 Besides text data,
there are also  Data Visualization
methods to visualize
relationships, such as
visualizing social
 Measuring Data Similarity and Dissimilarity
networks
 Summary

Newsmap: Google News Stories in 2005

48

8
 Similarity and Dissimilarity Data Matrix and Dissimilarity Matrix
 Similarity  Data matrix
 Numerical measure of how alike two data objects are  n data points with p  x 11 ... x 1f ... x 1p 
 
 Value is higher when objects are more alike dimensions  ... ... ... ... ... 
x x ip 
 Two modes
... x if ...
 Often falls in the range [0,1]  i1 
 ... ... ... ... ... 
Dissimilarity (e.g., distance) x x np 

... x nf ...
 Numerical measure of how different two data objects
 n1 
are  Dissimilarity matrix
 0 
 Lower when objects are more alike  n data points, but
 d(2,1) 0 
 Minimum dissimilarity is often 0 registers only the  
 d(3,1 ) 
 Upper limit varies
distance 
d ( 3,2 ) 0

 A triangular matrix  : : : 
 Proximity refers to a similarity or dissimilarity  d ( n ,1) d ( n ,2 ) ... ... 0 
 Single mode

49 50

Proximity Measure for Nominal Attributes Proximity Measure for Binary Attributes
Object j
 A contingency table for binary data
 Can take 2 or more states, e.g., red, yellow, blue, Object i
green (generalization of a binary attribute)  Distance measure for symmetric
 Method 1: Simple matching binary variables:
Distance measure for asymmetric
m: # of matches, p: total # of variables


binary variables:
d ( i , j )  p p m  Jaccard coefficient (similarity

 Method 2: Use a large number of binary attributes measure for asymmetric binary
variables):
 creating a new binary attribute for each of the  Note: Jaccard coefficient is the same as “coherence”:
M nominal states

51 52

Dissimilarity between Binary Variables Standardizing Numeric Data


z  x
 Z-score:
 Example
 X: raw score to be standardized, μ: mean of the population, σ:
Name Gender Fever Cough Test-1 Test-2 Test-3 Test-4 standard deviation
Jack M Y N P N N N  the distance between the raw score and the population mean in
Mary F Y N P N P N
units of the standard deviation
Jim M Y P N N N N
 negative when the raw score is below the mean, “+” when above
 Gender is a symmetric attribute  An alternative way: Calculate the mean absolute deviation
 The remaining attributes are asymmetric binary s f  1n (| x1 f  m f |  | x2 f  m f | ... | xnf  m f |)
Let the values Y and P be 1, and the value N 0 where

m f  1n (x1 f  x2 f  ...  xnf )
x m
.

0  1
d ( jack , mary ) 
2  0  1
 0 . 33 zif  if
sf
f

1  1
 standardized measure (z-score):
d ( jack , jim )   0 . 67
1  1  1  Using mean absolute deviation is more robust than using standard
d ( jim , mary ) 
1  2
 0 . 75 deviation
1  1  2
53 54

9
 Example:
Data Matrix and Dissimilarity Matrix Distance on Numeric Data: Minkowski Distance
 Minkowski distance: A popular distance measure
x2 x4
Data Matrix
point attribute1 attribute2
4 x1 1 2
x2 3 5
x3 2 0 where i = (xi1, xi2, …, xip) and j = (xj1, xj2, …, xjp) are two
x4 4 5 p-dimensional data objects, and h is the order (the
2 x1 distance so defined is also called L-h norm)
Dissimilarity Matrix  Properties
(with Euclidean Distance)
x3  d(i, j) > 0 if i ≠ j, and d(i, i) = 0 (Positive definiteness)
x1 x2 x3 x4
d(i, j) = d(j, i) (Symmetry)
0 2 4
x1 0 

d(i, j)  d(i, k) + d(k, j) (Triangle Inequality)

x2 3.61 0

x3 5.1 5.1 0
x4 4.24 1 5.39 0
 A distance that satisfies these properties is a metric
55 56

Special Cases of Minkowski Distance Example: Minkowski Distance

Dissimilarity Matrices
point attribute 1 attribute 2 Manhattan (L1)
 h = 1: Manhattan (city block, L1 norm) distance x1 1 2 L x1 x2 x3 x4
 E.g., the Hamming distance: the number of bits that are x2 3 5 x1 0
different between two binary vectors x3 2 0 x2 5 0
x4 4 5 x3 3 6 0
d (i, j) | x  x |  | x  x | ... | x  x |
i1 j1 i2 j 2 ip jp x4 6 1 7 0

 h = 2: (L2 norm) Euclidean distance Euclidean (L2)

L2 x1 x2 x3 x4
d (i, j)  (| x  x |2  | x  x |2 ... | x  x |2 ) x1 0
i1 j1 i2 j 2 ip jp
x2 3.61 0
 h  . “supremum” (Lmax norm, L norm) distance. x3 2.24 5.1 0
x4 4.24 1 5.39 0
 This is the maximum difference between any component

(attribute) of the vectors Supremum

L x1 x2 x3 x4
x1 0
x2 3 0
x3 2 5 0
x4 3 1 5 0
57 58

Ordinal Variables Attributes of Mixed Type

 A database may contain all attribute types

 An ordinal variable can be discrete or continuous  Nominal, symmetric binary, asymmetric binary, numeric,
 Order is important, e.g., rank ordinal
 Can be treated like interval-scaled  One may use a weighted formula to combine their effects
 replace xif by their rank r if  {1,..., M f }  pf  1 ij( f ) dij( f )
d (i, j) 
 map the range of each variable onto [0, 1] by replacing  pf  1 ij( f )
i-th object in the f-th variable by
 f is binary or nominal:
r if  1 dij(f) = 0 if xif = xjf , or dij(f) = 1 otherwise
z 
if M f  1  f is numeric: use the normalized distance
 compute the dissimilarity using methods for interval-  f is ordinal
scaled variables  Compute ranks rif and
zif  r  1 if

 Treat zif as interval-scaled M 1 f

59 60

10
 Cosine Similarity Example: Cosine Similarity
 A document can be represented by thousands of attributes, each  cos(d1, d2) = (d1  d2) /||d1|| ||d2|| ,
recording the frequency of a particular word (such as keywords) or where  indicates vector dot product, ||d|: the length of vector d
phrase in the document.
 Ex: Find the similarity between documents 1 and 2.

d1 = (5, 0, 3, 0, 2, 0, 0, 2, 0, 0)
d2 = (3, 0, 2, 0, 1, 1, 0, 1, 0, 1)

 Other vector objects: gene features in micro-arrays, … d1d2 = 5*3+0*0+3*2+0*0+2*1+0*1+0*1+2*1+0*0+0*1 = 25

 Applications: information retrieval, biologic taxonomy, gene feature ||d1||= (5*5+0*0+3*3+0*0+2*2+0*0+0*0+2*2+0*0+0*0)0.5=(42)0.5
mapping, ... = 6.481
 Cosine measure: If d1 and d2 are two vectors (e.g., term-frequency ||d2||= (3*3+0*0+2*2+0*0+1*1+1*1+0*0+1*1+0*0+1*1)0.5=(17)0.5
vectors), then = 4.12
cos(d1, d2) = (d1  d2) /||d1|| ||d2|| , cos(d1, d2 ) = 0.94
where  indicates vector dot product, ||d||: the length of vector d

61 62

Chapter 2: Getting to Know Your Data Summary

 Data attribute types: nominal, binary, ordinal, interval-scaled, ratio-
scaled
 Data Objects and Attribute Types
 Many types of data sets, e.g., numerical, text, graph, Web, image.
 Gain insight into the data by:
 Basic Statistical Descriptions of Data
 Basic statistical data description: central tendency, dispersion,
graphical displays
 Data Visualization  Data visualization: map data onto graphical primitives
 Measure data similarity
 Measuring Data Similarity and Dissimilarity  Above steps are the beginning of data preprocessing.
 Many methods have been developed but still an active area of research.

 Summary

63 64

References
 W. Cleveland, Visualizing Data, Hobart Press, 1993
 T. Dasu and T. Johnson. Exploratory Data Mining and Data Cleaning. John Wiley, 2003
 U. Fayyad, G. Grinstein, and A. Wierse. Information Visualization in Data Mining and
Knowledge Discovery, Morgan Kaufmann, 2001
 L. Kaufman and P. J. Rousseeuw. Finding Groups in Data: an Introduction to Cluster
Analysis. John Wiley & Sons, 1990.
 H. V. Jagadish, et al., Special Issue on Data Reduction Techniques. Bulletin of the Tech.
Committee on Data Eng., 20(4), Dec. 1997
 D. A. Keim. Information visualization and visual data mining, IEEE trans. on
Visualization and Computer Graphics, 8(1), 2002
 D. Pyle. Data Preparation for Data Mining. Morgan Kaufmann, 1999
 S. Santini and R. Jain,” Similarity measures”, IEEE Trans. on Pattern Analysis and
Machine Intelligence, 21(9), 1999
 E. R. Tufte. The Visual Display of Quantitative Information, 2nd ed., Graphics Press,
2001
 C. Yu , et al., Visual data mining of multimedia data for social and behavioral studies,
Information Visualization, 8(1), 2009
65

11