Data Mining Applications of Cluster Analysis

Cluster Analysis: Basic Concepts

and Algorithms ! Understanding Discovered Clusters Industry Group
Applied-Matl-DOWN,Bay-Network-Down,3-COM-DOWN,

– Group related documents

1 Cabletron-Sys-DOWN,CISCO-DOWN,HP-DOWN,
DSC-Comm-DOWN,INTEL-DOWN,LSI-Logic-DOWN,
Micron-Tech-DOWN,Texas-Inst-Down,Tellabs-Inc-Down,
Technology1-DOWN
Natl-Semiconduct-DOWN,Oracl-DOWN,SGI-DOWN,
for browsing, group genes Sun-DOWN
Apple-Comp-DOWN,Autodesk-DOWN,DEC-DOWN,

and proteins that have 2 ADV-Micro-Device-DOWN,Andrew-Corp-DOWN,

Computer-Assoc-DOWN,Circuit-City-DOWN,
Compaq-DOWN, EMC-Corp-DOWN, Gen-Inst-DOWN,
Technology2-DOWN
similar functionality, or Motorola-DOWN,Microsoft-DOWN,Scientific-Atl-DOWN
Fannie-Mae-DOWN,Fed-Home-Loan-DOWN,
group stocks with similar 3 MBNA-Corp-DOWN,Morgan-Stanley-DOWN Financial-DOWN

price fluctuations Baker-Hughes-UP,Dresser-Inds-UP,Halliburton-HLD-UP,

4 Louisiana-Land-UP,Phillips-Petro-UP,Unocal-UP,
Schlumberger-UP
Oil-UP

! Summarization
– Reduce the size of large
data sets

Clustering precipitation
in Australia

What is Cluster Analysis? What is not Cluster Analysis?

! Finding groups of objects such that the objects in a group ! Supervised classification
will be similar (or related) to one another and different – Have class label information
from (or unrelated to) the objects in other groups
! Simple segmentation
Inter-cluster
– Dividing students into different registration groups
Intra-cluster distances are alphabetically, by last name
distances are maximized
minimized
! Results of a query
– Groupings are a result of an external specification

! Graph partitioning
– Some mutual relevance and synergy, but areas are not
identical
Notion of a Cluster can be Ambiguous Partitional Clustering

How many clusters? Six Clusters

Two Clusters Four Clusters

Original Points A Partitional Clustering

Types of Clusterings Hierarchical Clustering

! A clustering is a set of clusters

! Important distinction between hierarchical and p1

p3 p4
partitional sets of clusters p2
p1 p2 p3 p4
! Partitional Clustering Traditional Hierarchical Clustering Traditional Dendrogram

– A division data objects into non-overlapping subsets (clusters)

such that each data object is in exactly one subset

! Hierarchical clustering p1
p3 p4
– A set of nested clusters organized as a hierarchical tree p2
p1 p2 p3 p4
Non-traditional Hierarchical Clustering Non-traditional Dendrogram
Other Distinctions Between Sets of Clusters Types of Clusters: Well-Separated

! Exclusive versus non-exclusive ! Well-Separated Clusters:

– In non-exclusive clusterings, points may belong to multiple – A cluster is a set of points such that any point in a cluster is
clusters. closer (or more similar) to every other point in the cluster than
– Can represent multiple classes or border points to any point not in the cluster.
! Fuzzy versus non-fuzzy
– In fuzzy clustering, a point belongs to every cluster with some
weight between 0 and 1
– Weights must sum to 1
– Probabilistic clustering has similar characteristics
! Partial versus complete
– In some cases, we only want to cluster some of the data
! Heterogeneous versus homogeneous
– Cluster of widely different sizes, shapes, and densities
3 well-separated clusters

Types of Clusters Types of Clusters: Center-Based

! Well-separated clusters ! Center-based

– A cluster is a set of objects such that an object in a cluster is
closer (more similar) to the center of a cluster, than to the
! Center-based clusters center of any other cluster
– The center of a cluster is often a centroid, the average of all
the points in the cluster, or a medoid, the most
! Contiguous clusters representative point of a cluster

! Density-based clusters

! Property or Conceptual

! Described by an Objective Function 4 center-based clusters

Types of Clusters: Contiguity-Based Types of Clusters: Conceptual Clusters

! Contiguous Cluster (Nearest neighbor or ! Shared Property or Conceptual Clusters

Transitive) – Finds clusters that share some common property or represent
– A cluster is a set of points such that a point in a cluster is a particular concept.
closer (or more similar) to one or more other points in the .
cluster than to any point not in the cluster.

8 contiguous clusters 2 Overlapping Circles

Types of Clusters: Density-Based Types of Clusters: Objective Function

! Density-based ! Clusters Defined by an Objective Function

– A cluster is a dense region of points, which is separated by – Finds clusters that minimize or maximize an objective function.
low-density regions, from other regions of high density.
– Enumerate all possible ways of dividing the points into clusters and
– Used when the clusters are irregular or intertwined, and when evaluate the `goodness' of each potential set of clusters by using
noise and outliers are present.
the given objective function. (NP Hard)
– Can have global or local objectives.
! Hierarchical clustering algorithms typically have local objectives
! Partitional algorithms typically have global objectives
– A variation of the global objective function approach is to fit the
data to a parameterized model.
! Parameters for the model are determined from the data.
! Mixture models assume that the data is a mixture' of a number of
statistical distributions.
6 density-based clusters
Types of Clusters: Objective Function … Clustering Algorithms

! Map the clustering problem to a different domain ! K-means and its variants
and solve a related problem in that domain
– Proximity matrix defines a weighted graph, where the ! Hierarchical clustering
nodes are the points being clustered, and the
weighted edges represent the proximities between
! Density-based clustering
points

– Clustering is equivalent to breaking the graph into

connected components, one for each cluster.

– Want to minimize the edge weight between clusters

and maximize the edge weight within clusters

Characteristics of the Input Data Are Important K-means Clustering

! Type of proximity or density measure ! Partitional clustering approach

– This is a derived measure, but central to clustering ! Each cluster is associated with a centroid (center point)
! Sparseness ! Each point is assigned to the cluster with the closest
– Dictates type of similarity centroid
– Adds to efficiency
! Number of clusters, K, must be specified
! Attribute type ! The basic algorithm is very simple
– Dictates type of similarity
! Type of Data
– Dictates type of similarity
– Other characteristics, e.g., autocorrelation
! Dimensionality
! Noise and Outliers
! Type of Distribution
K-means Clustering – Details Importance of Choosing Initial Centroids
! Initial centroids are often chosen randomly.
Iteration 6
1
2
3
4
5
– Clusters produced vary from one run to another. 3

! The centroid is (typically) the mean of the points in the

2.5
cluster.
! Closeness is measured by Euclidean distance, cosine 2

similarity, correlation, etc. 1.5

K-means will converge for common similarity measures

y
!
mentioned above. 1

! Most of the convergence happens in the first few 0.5

iterations. 0
– Often the stopping condition is changed to Until relatively few
points change clusters -2 -1.5 -1 -0.5 0 0.5 1 1.5 2
x
! Complexity is O( n * K * I * d )
– n = number of points, K = number of clusters,
I = number of iterations, d = number of attributes

Two different K-means Clusterings Importance of Choosing Initial Centroids

3
Iteration 1 Iteration 2 Iteration 3
3 3 3
2.5
2.5 2.5 2.5
2
Original Points 2 2 2
1.5
y

1.5 1.5 1.5

1

y
1 1 1
0.5
0.5 0.5 0.5
0
0 0 0

-2 -1.5 -1 -0.5 0 0.5 1 1.5 2

x -2 -1.5 -1 -0.5 0 0.5 1 1.5 2 -2 -1.5 -1 -0.5 0 0.5 1 1.5 2 -2 -1.5 -1 -0.5 0 0.5 1 1.5 2
x x x

3 3
Iteration 4 Iteration 5 Iteration 6
3 3 3
2.5 2.5

2.5 2.5 2.5

2 2

2 2 2
1.5 1.5
y

1.5 1.5 1.5

1 1
y

y
1 1 1
0.5 0.5

0 0 0.5 0.5 0.5

0 0 0
-2 -1.5 -1 -0.5 0 0.5 1 1.5 2 -2 -1.5 -1 -0.5 0 0.5 1 1.5 2
x x
-2 -1.5 -1 -0.5 0 0.5 1 1.5 2 -2 -1.5 -1 -0.5 0 0.5 1 1.5 2 -2 -1.5 -1 -0.5 0 0.5 1 1.5 2

Optimal Clustering Sub-optimal Clustering x x x

Evaluating K-means Clusters Importance of Choosing Initial Centroids …

! Most common measure is Sum of Squared Error (SSE) 3

Iteration 1
3
Iteration 2

– For each point, the error is the distance to the nearest cluster
2.5 2.5

2 2

– To get SSE, we square these errors and sum them. 1.5 1.5

y
K 1 1

SSE = ! ! dist 2 (mi , x ) 0.5

0
0.5

i =1 x"Ci
-2 -1.5 -1 -0.5 0 0.5 1 1.5 2 -2 -1.5 -1 -0.5 0 0.5 1 1.5 2

– x is a data point in cluster Ci and mi is the representative point for

x x

cluster Ci Iteration 3 Iteration 4 Iteration 5

3 3 3

! can show that mi corresponds to the center (mean) of the cluster 2.5 2.5 2.5

– Given two clusters, we can choose the one with the smallest 2 2 2

error 1.5 1.5 1.5

y
– One easy way to reduce SSE is to increase K, the number of 1 1 1

clusters 0.5 0.5 0.5

0 0 0

! A good clustering with smaller K can have a lower SSE than a poor
clustering with higher K -2 -1.5 -1 -0.5 0
x
0.5 1 1.5 2 -2 -1.5 -1 -0.5 0
x
0.5 1 1.5 2 -2 -1.5 -1 -0.5 0
x
0.5 1 1.5 2

Importance of Choosing Initial Centroids … Problems with Selecting Initial Points

Iteration 5
1
2
3
4 ! If there are K real clusters then the chance of selecting
3 one centroid from each cluster is small.
2.5
– Chance is relatively small when K is large
– If clusters are the same size, n, then
2

1.5
y

0.5
– For example, if K = 10, then probability = 10!/1010 = 0.00036
0
– Sometimes the initial centroids will readjust themselves in
right way, and sometimes they don t
-2 -1.5 -1 -0.5 0 0.5 1 1.5 2
x – Consider an example of five pairs of clusters
10 Clusters Example 10 Clusters Example
Iteration 4
1
2
3 Iteration 4
1
2
3
8 8

6 6

4 4

2 2
y

y
0 0

-2 -2

-4 -4

-6 -6

0 5 10 15 20 0 5 10 15 20
x x
Starting with two initial centroids in one cluster of each pair of clusters Starting with some pairs of clusters having three initial centroids, while other have only one.

10 Clusters Example 10 Clusters Example

Iteration 1 Iteration 2
Iteration 1 Iteration 2 8 8
8 8

6 6
6 6

4 4
4 4

2 2
2 2

y
0 0
y

0 0

-2 -2
-2 -2

-4 -4
-4 -4

-6 -6
-6 -6

0 5 10 15 20 0 5 10 15 20
0 5 10 15 20 0 5 10 15 20
Iteration
x 3 Iteration
x 4
x x
Iteration 3 Iteration 4 8 8
8 8
6 6
6 6
4 4
4 4
2 2
2 2

y
0 0
y

0 0
-2 -2
-2 -2
-4 -4
-4 -4
-6 -6
-6 -6
0 5 10 15 20 0 5 10 15 20
0 5 10 15 20 0 5 10 15 20 x x
x x

Starting with two initial centroids in one cluster of each pair of clusters Starting with some pairs of clusters having three initial centroids, while other have only one.
Solutions to Initial Centroids Problem Updating Centers Incrementally

! Multiple runs ! In the basic K-means algorithm, centroids are

– Helps, but probability is not on your side updated after all points are assigned to a centroid
! Sample and use hierarchical clustering to
determine initial centroids ! An alternative is to update the centroids after
! Select more than k initial centroids and then each assignment (incremental approach)
select among these initial centroids – Each assignment updates zero or two centroids
– Select most widely separated – More expensive
! Postprocessing – Introduces an order dependency
! Bisecting K-means – Never get an empty cluster
– Not as susceptible to initialization issues – Can use weights to change the impact

Handling Empty Clusters Pre-processing and Post-processing

! Basic K-means algorithm can yield empty clusters ! Pre-processing

– Normalize the data
! Several strategies – Eliminate outliers
– Choose the point that contributes most to SSE
! Post-processing
– Choose a point from the cluster with the highest SSE
– Eliminate small clusters that may represent outliers
– If there are several empty clusters, the above can be
repeated several times. – Split loose clusters, i.e., clusters with relatively high
SSE
– Merge clusters that are close and that have
relatively low SSE
– Can use these steps during the clustering process
! ISODATA
Bisecting K-means Limitations of K-means

! Bisecting K-means algorithm ! K-means has problems when clusters are of

– Variant of K-means that can produce a partitional or a differing
hierarchical clustering
– Sizes
– Densities
– Non-globular shapes

! K-means has problems when the data contains

outliers.

Bisecting K-means Example Limitations of K-means: Differing Sizes

Original Points K-means (3 Clusters)

Limitations of K-means: Differing Density Overcoming K-means Limitations

Original Points K-means (3 Clusters) Original Points K-means Clusters

One solution is to use many clusters.

Find parts of clusters, but need to put together.

Limitations of K-means: Non-globular Shapes Overcoming K-means Limitations

Original Points K-means (2 Clusters) Original Points K-means Clusters

Overcoming K-means Limitations Strengths of Hierarchical Clustering

! Do not have to assume any particular number of

clusters
– Any desired number of clusters can be obtained by
cutting the dendogram at the proper level

! They may correspond to meaningful taxonomies

– Example in biological sciences (e.g., animal kingdom,
phylogeny reconstruction, !)

Original Points K-means Clusters

Hierarchical Clustering Hierarchical Clustering

! Produces a set of nested clusters organized as a ! Two main types of hierarchical clustering
hierarchical tree – Agglomerative:
! Start with the points as individual clusters
! Can be visualized as a dendrogram ! At each step, merge the closest pair of clusters until only one cluster
– A tree like diagram that records the sequences of (or k clusters) left
merges or splits
– Divisive:
6 5 ! Start with one, all-inclusive cluster
0.2
4 ! At each step, split a cluster until each cluster contains a point (or
0.15
3
2
4
there are k clusters)
5

0.1
2

0.05
1
1
! Traditional hierarchical algorithms use a similarity or
distance matrix
3

0
1 3 2 5 4 6

– Merge or split one cluster at a time

Agglomerative Clustering Algorithm Intermediate Situation

! More popular hierarchical clustering technique ! After some merging steps, we have some clusters
C1 C2 C3 C4 C5
! Basic algorithm is straightforward
C1
1. Compute the proximity matrix
C2
2. Let each data point be a cluster
C3
3. Repeat C3
C4
4. Merge the two closest clusters C4
C5
5. Update the proximity matrix
Proximity Matrix
6. Until only a single cluster remains C1

! Key operation is the computation of the proximity of

two clusters C2 C5

– Different approaches to defining the distance between

clusters distinguish the different algorithms ...
p1 p2 p3 p4 p9 p10 p11 p12

Starting Situation Intermediate Situation

! Start with clusters of individual points and a ! We want to merge the two closest clusters (C2 and C5) and
update the proximity matrix.
proximity matrix p1 p2 p3 p4 p5
C1 C2 C3 C4 C5
... C1
p1
C2
p2
C3
p3 C3
C4
p4 C4
C5
p5
.
Proximity Matrix
.
C1
. Proximity Matrix

C2 C5

... ...
p1 p2 p3 p4 p9 p10 p11 p12
p1 p2 p3 p4 p9 p10 p11 p12
After Merging How to Define Inter-Cluster Similarity
p1 p2 p3 p4 p5 ...
! The question is How do we update the proximity matrix?
p1
C2
U
C1 C5 C3 C4 p2

C1 ? p3

C2 U C5 ? ? ? ? p4
C3
C3 ? p5
C4 ! MIN
? .
C4 ! MAX
.
Proximity Matrix ! Group Average
C1 .
Proximity Matrix
! Distance Between Centroids
! Other methods driven by an objective
C2 U C5
function
– Ward s Method uses squared error
...
p1 p2 p3 p4 p9 p10 p11 p12

How to Define Inter-Cluster Similarity How to Define Inter-Cluster Similarity

p1 p2 p3 p4 p5 ... p1 p2 p3 p4 p5 ...
p1 p1
Similarity?
p2 p2

p3 p3

p4 p4

p5 p5
! MIN ! MIN
. .
! MAX ! MAX
. .
! Group Average .
! Group Average .
Proximity Matrix Proximity Matrix
! Distance Between Centroids ! Distance Between Centroids
! Other methods driven by an objective ! Other methods driven by an objective
function function
– Ward s Method uses squared error – Ward s Method uses squared error
How to Define Inter-Cluster Similarity Cluster Similarity: MIN or Single Link
p1 p2 p3 p4 p5 ...
p1
! Similarity of two clusters is based on the two most
p2
similar (closest) points in the different clusters
p3 – Determined by one pair of points, i.e., by one link in
the proximity graph.
p4

p5
! MIN
.
! MAX
. I1 I2 I3 I4 I5
! Group Average .
Proximity Matrix I1 1.00 0.90 0.10 0.65 0.20
! Distance Between Centroids I2 0.90 1.00 0.70 0.60 0.50
! Other methods driven by an objective I3 0.10 0.70 1.00 0.40 0.30
function I4 0.65 0.60 0.40 1.00 0.80
– Ward s Method uses squared error I5 0.20 0.50 0.30 0.80 1.00 1 2 3 4 5

How to Define Inter-Cluster Similarity Hierarchical Clustering: MIN

p1 p2 p3 p4 p5 ...
p1
5
! ! p2 1
3
p3

p4
5 0.2

2 1 0.15
p5
! MIN 2 3 6
. 0.1
! MAX
. 0.05
! Group Average . 4
Proximity Matrix 4
! Distance Between Centroids 0
3 6 2 5 4 1

! Other methods driven by an objective

function
– Ward s Method uses squared error Nested Clusters Dendrogram
Strength of MIN Cluster Similarity: MAX or Complete Linkage

! Similarity of two clusters is based on the two least

similar (most distant) points in the different
clusters
– Determined by all pairs of points in the two clusters

I1 I2 I3 I4 I5
I1 1.00 0.90 0.10 0.65 0.20
Original Points Two Clusters
I2 0.90 1.00 0.70 0.60 0.50
I3 0.10 0.70 1.00 0.40 0.30
I4 0.65 0.60 0.40 1.00 0.80
• Can handle non-elliptical shapes I5 0.20 0.50 0.30 0.80 1.00 1 2 3 4 5

Limitations of MIN Hierarchical Clustering: MAX

4 1
2 5 0.4

0.35
5
2 0.3

0.25

3 6
0.2

3 0.15
1 0.1

4 0.05

0
3 6 4 1 2 5

Original Points Two Clusters

Nested Clusters Dendrogram

• Sensitive to noise and outliers
Strength of MAX Cluster Similarity: Group Average

! Proximity of two clusters is the average of pairwise proximity

between points in the two clusters.
" proximity(p , p )
pi#Clusteri
i j

p j#Clusterj
proximity(Clusteri , Clusterj ) =
|Clusteri |!|Clusterj |

! Need to use average connectivity for scalability since total

proximity favors large clusters
I1 I2 I3 I4 I5
Original Points Two Clusters I1 1.00 0.90 0.10 0.65 0.20
I2 0.90 1.00 0.70 0.60 0.50
I3 0.10 0.70 1.00 0.40 0.30
• Less susceptible to noise and outliers I4 0.65 0.60 0.40 1.00 0.80
I5 0.20 0.50 0.30 0.80 1.00 1 2 3 4 5

Limitations of MAX Hierarchical Clustering: Group Average

5 4 1
2
0.25

5 0.2
2
0.15

3 6 0.1

1 0.05

4 0
3 3 6 4 1 2 5

Original Points Two Clusters

•Tends to break large clusters Nested Clusters Dendrogram

•Biased towards globular clusters
Hierarchical Clustering: Group Average Hierarchical Clustering: Comparison

! Compromise between Single and Complete 3

1
5
4 1
Link 5
2 5
2 5
1 2
MIN MAX
2 3 6 3 6
3
1
! Strengths 4
4 4

– Less susceptible to noise and outliers

5
1 5 4 1
2 2
! Limitations 5
2 Ward s Method 5
2
Group Average
– Biased towards globular clusters 3
3 6 3 6
4 1 1
4 4
3

Cluster Similarity: Ward s Method Hierarchical Clustering: Time and Space requirements

! Similarity of two clusters is based on the increase ! O(N2) space since it uses the proximity matrix.
in squared error when two clusters are merged – N is the number of points.
– Similar to group average if distance between points is
distance squared
! O(N3) time in many cases
– There are N steps and at each step the size, N2,
! Less susceptible to noise and outliers proximity matrix must be updated and searched
– Complexity can be reduced to O(N2 log(N) ) time for
! Biased towards globular clusters some approaches

! Hierarchical analogue of K-means

– Can be used to initialize K-means
Hierarchical Clustering: Problems and Limitations MST: Divisive Hierarchical Clustering

! Once a decision is made to combine two clusters, ! Use MST for constructing hierarchy of clusters
it cannot be undone

! No objective function is directly minimized

! Different schemes have problems with one or

more of the following:
– Sensitivity to noise and outliers
– Difficulty handling different sized clusters and convex
shapes
– Breaking large clusters

MST: Divisive Hierarchical Clustering DBSCAN

! Build MST (Minimum Spanning Tree) ! DBSCAN is a density-based algorithm.

– Start with a tree that consists of any point – Density = number of points within a specified radius (Eps)
– In successive steps, look for the closest pair of points (p, q) such
that one point (p) is in the current tree but the other (q) is not – A point is a core point if it has more than a specified number
– Add q to the tree and put an edge between p and q of points (MinPts) within Eps
! These are points that are at the interior of a cluster

– A border point has fewer than MinPts within Eps, but is in

the neighborhood of a core point

– A noise point is any point that is not a core point or a border

point.
DBSCAN: Core, Border, and Noise Points DBSCAN: Core, Border and Noise Points

Original Points Point types: core,

border and noise

Eps = 10, MinPts = 4

DBSCAN Algorithm When DBSCAN Works Well

! Eliminate noise points

! Perform clustering on the remaining points

Original Points Clusters

• Resistant to Noise
• Can handle clusters of different shapes and sizes
When DBSCAN Does NOT Work Well Cluster Validity

! For supervised classification we have a variety of

measures to evaluate how good our model is
– Accuracy, precision, recall

! For cluster analysis, the analogous question is how to

evaluate the goodness of the resulting clusters?
(MinPts=4, Eps=9.75).
! But clusters are in the eye of the beholder !
Original Points
! Then why do we want to evaluate them?
– To avoid finding patterns in noise
– To compare clustering algorithms
• Varying densities
– To compare two sets of clusters
– To compare two clusters
• High-dimensional data
(MinPts=4, Eps=9.92)

DBSCAN: Determining EPS and MinPts Clusters found in Random Data

1 1

! Idea is that for points in a cluster, their kth nearest 0.9 0.9

neighbors are at roughly the same distance 0.8

0.7
0.8

0.7

! Noise points have the kth nearest neighbor at farther Random

Points
0.6

0.5
0.6

0.5
DBSCAN

y
distance

y
0.4 0.4

0.3 0.3

! So, plot sorted distance of every point to its kth 0.2 0.2

nearest neighbor 0.1

0
0.1

0
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
x x
1 1

0.9 0.9

K-means 0.8 0.8

Complete
0.7 0.7
Link
0.6 0.6

0.5 0.5
y

y
0.4 0.4

0.3 0.3

0.2 0.2

0.1 0.1

0 0
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
x x
Different Aspects of Cluster Validation Measuring Cluster Validity Via Correlation

1. Determining the clustering tendency of a set of data, i.e., ! Two matrices

distinguishing whether non-random structure actually exists in the – Proximity Matrix
data. – Incidence Matrix
2. Comparing the results of a cluster analysis to externally known ! One row and one column for each data point
results, e.g., to externally given class labels. ! An entry is 1 if the associated pair of points belong to the same cluster
3. Evaluating how well the results of a cluster analysis fit the data ! An entry is 0 if the associated pair of points belongs to different clusters
without reference to external information. ! Compute the correlation between the two matrices
- Use only the data
– Since the matrices are symmetric, only the correlation between
4. Comparing the results of two different sets of cluster analyses to n(n-1) / 2 entries needs to be calculated.
determine which is better.
! High correlation indicates that points that belong to the
5. Determining the correct number of clusters. same cluster are close to each other.
! Not a good measure for some density or contiguity based
For 2, 3, and 4, we can further distinguish whether we want to clusters.
evaluate the entire clustering or just individual clusters.

Measures of Cluster Validity Measuring Cluster Validity Via Correlation

! Numerical measures that are applied to judge various aspects

of cluster validity, are classified into the following three types. ! Correlation of incidence and proximity matrices
– External Index: Used to measure the extent to which cluster labels for the K-means clusterings of the following two
match externally supplied class labels. data sets.
! Entropy
– Internal Index: Used to measure the goodness of a clustering 1

0.9
1

0.9
structure without respect to external information. 0.8 0.8

! Sum of Squared Error (SSE) 0.7 0.7

0.6 0.6
– Relative Index: Used to compare two different clusterings or 0.5 0.5

y
clusters. 0.4 0.4

! Often an external or internal index is used for this function, e.g., SSE or entropy 0.3 0.3

0.2 0.2

! Sometimes these are referred to as criteria instead of indices 0.1 0.1

0 0
– However, sometimes criterion is the general strategy and index is the 0 0.2 0.4
x
0.6 0.8 1 0 0.2 0.4
x
0.6 0.8 1

numerical measure that implements the criterion.

Corr = -0.9235 Corr = -0.5810

Using Similarity Matrix for Cluster Validation Using Similarity Matrix for Cluster Validation

! Order the similarity matrix with respect to cluster

! Clusters in random data are not so crisp
labels and inspect visually.
1 1

1 10 0.9 0.9
1
10 20 0.8 0.8
0.9
0.9
30 0.7 0.7
20 0.8
0.8 40 0.6
0.6
30

Points
0.7
0.7 50 0.5 0.5

y
40 0.6
0.6 60 0.4 0.4

Points
50 0.5 70 0.3 0.3
0.5
y

60 0.4 80 0.2 0.2

0.4
90 0.1 0.1
70 0.3
0.3
100 0 0
80 0.2 20 40 60 80 100 Similarity 0 0.2 0.4 0.6 0.8 1
0.2 x
Points
90 0.1
0.1
100 0
0 20 40 60 80 100 Similarity
0 0.2 0.4 0.6 0.8 1
Points
x
K-means

Using Similarity Matrix for Cluster Validation Using Similarity Matrix for Cluster Validation

! Clusters in random data are not so crisp ! Clusters in random data are not so crisp

1 1 1 1

10 0.9 0.9 10 0.9 0.9

20 0.8 0.8 20 0.8 0.8

30 0.7 0.7 30 0.7 0.7

40 0.6 0.6 40 0.6 0.6

Points
Points

50 0.5 0.5 50 0.5 0.5

y
y

60 0.4 0.4 60 0.4 0.4

70 0.3 0.3 70 0.3 0.3

80 0.2 0.2 80 0.2 0.2

90 0.1 0.1 90 0.1 0.1

100 0 0 100 0 0
20 40 60 80 100 Similarity 0 0.2 0.4 0.6 0.8 1 20 40 60 80 100 Similarity 0 0.2 0.4 0.6 0.8 1
Points x Points x

DBSCAN Complete Link

Using Similarity Matrix for Cluster Validation Internal Measures: SSE

! SSE curve for a more complicated data set

1

0.9
1 500
2 0.8
6
0.7
1000
3 0.6 1
4 2 6
1500 0.5

0.4 3
4
2000
0.3
5
0.2
2500
0.1
7 5
3000 0
500 1000 1500 2000 2500 3000
7

DBSCAN
SSE of clusters found using K-means

Internal Measures: SSE Framework for Cluster Validity

! Clusters in more complicated figures aren t well separated ! Need a framework to interpret any measure.
! Internal Index: Used to measure the goodness of a clustering – For example, if our measure of evaluation has the value, 10, is that
structure without respect to external information good, fair, or poor?

– SSE ! Statistics provide a framework for cluster validity

– The more atypical a clustering result is, the more likely it represents
! SSE is good for comparing two clusterings or two clusters valid structure in the data
(average SSE). – Can compare the values of an index that result from random data or
! Can also be used to estimate the number of clusters clusterings to those of a clustering result.
10

9
! If the value of the index is unlikely, then the cluster results are valid
6
8 – These approaches are more complicated and harder to understand.
4
7

2 6 ! For comparing the results of two different sets of cluster

SSE

analyses, a framework is less necessary.

5
0
4
-2 3
– However, there is the question of whether the difference between two
2
-4
1
index values is significant
-6 0
2 5 10 15 20 25 30
5 10 15
K
 Statistical Framework for SSE Internal Measures: Cohesion and Separation

! Example ! Cluster Cohesion: Measures how closely related

– Compare SSE of 0.005 against three clusters in random data are objects in a cluster
– Histogram shows SSE of three clusters in 500 sets of random data – Example: SSE
points of size 100 distributed over the range 0.2 – 0.8 for x and y
! Cluster Separation: Measure how distinct or well-
values
separated a cluster is from other clusters
1

Example: Squared Error

50
0.9
45
!
0.8
40 – Cohesion is measured by the within cluster sum of squares (SSE)
WSS = ! ! ( x # mi )2
0.7
35
0.6
30

i x"C i
Count
0.5
y

25
0.4
20
0.3
15
– Separation is measured by the between cluster sum of squares
0.2
10
0.1

0
5 BSS = ! Ci (m " mi )2
0 0.2 0.4 0.6 0.8 1 0
0.016 0.018 0.02 0.022 0.024 0.026 0.028 0.03 0.032 0.034 i
x SSE – Where |Ci| is the size of cluster i

Statistical Framework for Correlation Internal Measures: Cohesion and Separation

! Correlation of incidence and proximity matrices for the ! Example: SSE

K-means clusterings of the following two data sets. – BSS + WSS = constant
m
1 1 ! ! !
0.9 0.9
1 m1 2 3 4 m2 5
0.8 0.8

0.7 0.7

0.6 0.6

0.5 0.5
K=1 cluster: WSS= (1 ! 3)2 + (2 ! 3)2 + (4 ! 3)2 + (5 ! 3)2 = 10
y

0.4 0.4

BSS= 4 " (3 ! 3)2 = 0

0.3 0.3

0.2 0.2

0.1 0.1

0
0 0.2 0.4 0.6 0.8 1
0
0 0.2 0.4 0.6 0.8 1 Total = 10 + 0 = 10
x x

K=2 clusters: WSS= (1 ! 1.5)2 + (2 ! 1.5)2 + (4 ! 4.5)2 + (5 ! 4.5)2 = 1

Corr = -0.9235 Corr = -0.5810
BSS= 2 " (3 ! 1.5)2 + 2 " (4.5 ! 3)2 = 9
Total = 1 + 9 = 10
Internal Measures: Cohesion and Separation External Measures of Cluster Validity: Entropy and Purity

! A proximity graph based approach can also be used for

cohesion and separation.
– Cluster cohesion is the sum of the weight of all links within a cluster.
– Cluster separation is the sum of the weights between nodes in the cluster
and nodes outside the cluster.

cohesion separation

Internal Measures: Silhouette Coefficient Final Comment on Cluster Validity

! Silhouette Coefficient combine ideas of both cohesion and separation,

but for individual points, as well as clusters and clusterings The validation of clustering structures is the most
! For an individual point, i difficult and frustrating part of cluster analysis.
– Calculate a = average distance of i to the points in its cluster
– Calculate b = min (average distance of i to points in another cluster) Without a strong effort in this direction, cluster
– The silhouette coefficient for a point is then given by analysis will remain a black art accessible only to
s = 1 – a/b if a < b, (or s = b/a - 1 if a " b, not the usual case)
those true believers who have experience and
great courage.
b
– Typically between 0 and 1. a
– The closer to 1 the better.
Algorithms for Clustering Data, Jain and Dubes

! Can calculate the Average Silhouette width for a cluster or a

clustering