Sekizinci Ders
Sekizinci Ders
Kümeleme Analizi: Temel Kavramlar
Kümeleme Analizi: Temel Kavramlar
Dr. Hidayet Takçı
Dr. Hidayet Takçı
Kümeleme Analizi Nedir?
Kümeleme Analizi Nedir?
H
Heerr bbiirrii bbiirr ddiizzii öözznniitteelliikk iillee,, vveerrii nnookkttaallaarrıınnıınn bbiirr kküüm
meessii vvee
nnookkta
tala
larr aara
rası
sınd
ndaaki
ki bben
enzzeerl
rlii
ğ
ğ
ii ööllççeenn bbiirr bbeennzzeerrlliikk ööllççüüm
müü
verilmi
verilmi
şş
oollssuunn,, kküüm
meelleem
meenniinn aam
maaccıı;; aa
şş
aa
ğ
ğ
ıd
ıdak
akii öözzel
ellilikl
kleeri
ri
sa
sa
ğ
ğ
la
laya
yann kü
küme
mele
leriri bu
bulm
lmak
aktıtır.
r.
Kümeler arası
Kümeler arası
uzaklıklar
uzaklıklar
Küme içi
Küme içi
m
maa ss m
m zze
e ee
rr
u
uzzaa ıı aarr
minimize edilir
minimize edilir
Kümeleme Analizi Ne Değildir?
Kümeleme Analizi Ne Değildir?
Denetimli sınıflandırma
Denetimli sınıflandırma
–
–
Sınıf etiketi bilgisine sahip
Sınıf etiketi bilgisine sahip
Basit bölümleme
Basit bölümleme
–
–
Soyadına göre farklı kayıt gruplarının alfabetik olarak
Soyadına göre farklı kayıt gruplarının alfabetik olarak
bölünmesi
bölünmesi
Bir sorgunun sonuçları
Bir sorgunun sonuçları
–
–
Bir
Bir
şş
arta göre gruplamaların elde edilmesi
arta göre gruplamaların elde edilmesi
Kümeleme tipleri
Bir kümeleme kümelerin bir dizisidir.
Bölümlemeli ve hiyerar
ş
ik kümeler arasında
önemli bir ayrım vardır
Bölümlemeli Kümeleme
– Veri
nesnelerinin,
birbirini
kapsamayan
alt
kümelere
ayrılmasıdır. Her bir veri nesnesi altkümelerden sadece birinde
yer alır.
Hiyerar
ş
ik Kümeleme
Bölümlemeli Kümeleme
Hiyerarşik Kümeleme
p4 p1 p3 p2p4
p1 p2
p3
GelenekselHi erar ik
Kümeleme Geleneksel Dendro ramp4 p1 p3 p2
p4
p1 p2
p3
Geleneksel olmayanHiyerar
ş
ik
Kümeleme
Other Distinctions Between Sets of Clusters
Özel, özel olmayana kar
ş
ı
– Özel olmayan kümelemelerde noktalar birden fazla sınıfa ait
olabilirler.
– Çoklu sınıflar veya sınır noktaları sunulabilir mi?
Bulanık bulanık olmayana kar
ş
ı
– Bulanık sınıflandırmada bir nokta her bir kümeye 0 ve 1 aralı
ğ
ında bir
– A
ğ
ırlıklar toplamı 1 olmalıdır
–
htimali (Probabilistic) kümeleme ile benzer özellikleri vardır
Kısmi, bütüne kar
ş
ı
– Bazı durumlarda biz sadece verinin bir kısmı ile kümeleme yapmak
isteriz.
Heterojen, homojene kar
ş
ı
Kümelerin Tipleri
yi da
ğ
ıtılmı
ş
kümeler (Well-separated clusters)
Merkez tabanlı kümeler(Center-based clusters)
Biti
ş
ik Kümeler (Contiguous clusters)
Yo
ğ
unluk tabanlı kümeler (Density-based clusters)
Nitelik veya kavramsal (Property or Conceptual)
Bir amaç fonksiyonu tarafından açıklanan (Described by
Types of Clusters: Well-Separated
yi da
ğ
ıtılmı
ş
kümeler:
– Her bir nokta; kendi kümesindeki di
ğ
er noktalara daha yakın,
ba
ş
ka kümeden noktalara ise daha uzaktır. Böylesi kümeler iyi
da
ğ
ıtılmı
ş
kümelerdir.
Types of Clusters: Center-Based
Merkez tabanlı
– Küme içindeki bir nokta, kendi küme merkezine di
ğ
er küme
merkezlerine oranla daha yakın (veya daha benzer) ise bu
küme merkez tabanlı bir kümedir.
– Bir kümenin merkezi sıklıkla, ya kümedeki bütün noktaların bir
ortalaması olan centroid ile yada kümeyi sunmak için en uygun
nokta olan medoid ile sunulur.
Types of Clusters: Contiguity-Based
Biti
ş
ik küme (Nearest neighbor or Transitive)
– A cluster is a set of points such that a point in a cluster is
closer (or more similar) to one or more other points in the
cluster than to any point not in the cluster.
Types of Clusters: Density-Based
Yo
ğ
unluk tabanlı
– Daha dü
ş
ük yo
ğ
unluklu bölgelerden ayrılan daha yüksek
yo
ğ
unluklu noktaların bir kümesidir.
– Kümeler; düzensiz, birbirine karı
ş
mı
ş
veya gürültülü oldu
ğ
unda
kullanılır.
Types of Clusters: Conceptual Clusters
Payla
ş
ılan nitelik veya kavramsal kümeler
– Aynı ortak nitelikleri payla
ş
ır veya kısmi bir kavram sunar.
Types of Clusters: Objective Function
Bir amaç fonksiyonu tarafından tanımlaman
kümeler
– Bir amaç fonksiyonunu minimize veya maksimize eden kümeler
bulunur.
– Noktaların bölümlenmesi için olası bütün yollar sıralanır ve verilen
amaç fonksiyona göre kümelerin potansiyel dizilerinin en iyilikleri
de
ğ
erlendirilir (NP Hard)
–
Hiyerar
ş
ik kümeleme algoritmaları tipik olarak lokal amaç fonksiyonlara
sahiptir.
Bölümlemeli algoritmalar tipik olarak global amaç fonksiyonlara
sahiptir.
– Global amaç fonksiyonu yakla
ş
ımının bir çe
ş
idi veriyi parametrik
bir modele uydurmadır.
Modelin parametreleri veriden çıkarılır.
Mixture modeller veriyi birkaç istatistiksel da
ğ
ılımın bir karı
ş
ımı olarak
Kümeleme Algoritmaları
K-means ve onun çe
ş
itleri
Hiyerar
ş
ik kümeleme
K-means Kümeleme
Bölümlemeli kümeleme yakla
ş
ımıdır
Her bir küme bir centroid ile uyumludur (merkez nokta)
Her bir nokta kendisine en yakın centroid ile uyumlu
kümeye atanır
Kümelerin sayısı, K, belirlenmelidir
Temel al oritma ok basittir
K-means Kümeleme – Detaylar
Ba
ş
langıç merkez noktaları sıklıkla rastgele seçilir.
–
Kümeler bir çalı
ş
tırmadan di
ğ
erine de
ğ
i
ş
ebilir.
Centroid tipik olarak kümedeki noktaların bir ortalamasıdır.
‘Yakınlık’ Euclidean uzaklı
ğ
ı, cosine benzerli
ğ
i, correlation,
v.s. ile hesaplanabilir.
K-means yukarıdaki benzerlik ölçümlerini bir noktada bir
.
Hesaplamalar centroid sabit kalana kadar devam eder.
Karma
ş
ıklık O( n * K * I * d )
–
n = noktaların sayısı, K = kümelerin sayısı,
Importance of Choosing Initial Centroids
1.5 2 2.5 3 yIteration 1
1.5 2 2.5 3 yIteration 2
1.5 2 2.5 3 yIteration 3
1.5 2 2.5 3 yIteration 4
1.5 2 2.5 3 yIteration 5
1.5 2 2.5 3 yIteration 6
-2 -1.5 -1 -0.5 0 0.5 1 1.5 2 0 0.5 1x
-2 -1.5 -1 -0.5 0 0.5 1 1.5 2 0 0.5 1x
-2 -1.5 -1 -0.5 0 0.5 1 1.5 2 0 0.5 1x
-2 -1.5 -1 -0.5 0 0.5 1 1.5 2 0 0.5 1x
-2 -1.5 -1 -0.5 0 0.5 1 1.5 2 0 0.5 1x
-2 -1.5 -1 -0.5 0 0.5 1 1.5 2 0 0.5 1x
Importance of Choosing Initial Centroids
-2 -1.5 -1 -0.5 0 0.5 1 1.5 2 0 0.5 1 1.5 2 2.5 3 x y Iteration 1 -2 -1.5 -1 -0.5 0 0.5 1 1.5 2 0 0.5 1 1.5 2 2.5 3 x y Iteration 2 -2 -1.5 -1 -0.5 0 0.5 1 1.5 2 0 0.5 1 1.5 2 2.5 3 x y Iteration 3 -2 -1.5 -1 -0.5 0 0.5 1 1.5 2 0 0.5 1 1.5 2 2.5 3 x y Iteration 4 -2 -1.5 -1 -0.5 0 0.5 1 1.5 2 0 0.5 1 1.5 2 2.5 3 x y Iteration 5 -2 -1.5 -1 -0.5 0 0.5 1 1.5 2 0 0.5 1 1.5 2 2.5 3 x y Iteration 6K-means Kümelelerin Değerlendirmesi
En genel ölçüm hataların kareleri toplamıdır (Sum of
Squared Error (SSE))
– Her bir nokta için, hata en yakın kümeye olan uzaklıktır
– SSE hesabı için, bu hataların karesini hesaplar sonra toplarız.
∑ ∑
= ∈=
K i x C i ix
m
dist
SSE
1 2)
,
(
–
x , C
i
kümesinde bir veri noktasıdır ve
m
i,
C
i
kümesi için temsil
edici bir noktadır
m
i
küme için merkez noktayı temsil etmektedir.
–
ki küme verilmi
ş
olsun, biz bu iki küme için SSE de
ğ
erlerini
hesap eder ve en küçük olanı seçeriz.
– SSE de
ğ
erini azaltmak için bir yol K de
ğ
erini artırmaktır
Yüksek K ile zayıf kümelemeden daha dü
ş
ük SSE de
ğ
erine sahip
Boş Kümelerle Çalışma
Temel K-means algoritması bo
ş
kümelere sebep
olabilir
Birkaç strateji
– Choose the point that contributes most to SSE
– Choose a point from the cluster with the highest SSE
– If there are several empty clusters, the above can be
Updating Centers Incrementally
In the basic K-means algorithm, centroids are
updated after all points are assigned to a centroid
An alternative is to update the centroids after
each assignment (incremental approach)
– Each assignment updates zero or two centroids
– More expensive
– Introduces an order dependency
– Never get an empty cluster
Pre-processing and Post-processing
Pre-processing
– Normalize the data
– Eliminate outliers
Post-processing
–
– 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
Bisecting K-means
Bisecting K-means algorithm
–
Variant of K-means that can produce a partitional or a
Limitations of K-means
K-means has problems when clusters are of
differing
– Sizes
– Densities
– Non-globular shapes
K-means has problems when the data contains
Limitations of K-means: Differing Sizes
Limitations of K-means: Differing Density
Limitations of K-means: Non-globular Shapes
Overcoming K-means Limitations
Original Points
K-means Clusters
One solution is to use many clusters.
Overcoming K-means Limitations
Overcoming K-means Limitations
Hierarchical Clustering
Produces a set of nested clusters organized as a
hierarchical tree
Can be visualized as a dendrogram
– A tree like diagram that records the sequences of
merges or splits
1 3 2 5 4 6 0 0.05 0.1 0.15 0.2 1 2 3 4 5 6 1 2 3 4 5Strengths 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
– Example in biological sciences (e.g., animal kingdom,
phylogeny reconstruction, …)
Hierarchical Clustering
Two main types of hierarchical clustering
– Agglomerative:
Start with the points as individual clusters
At each step, merge the closest pair of clusters until only one cluster
(or k clusters) left
–
vsve:
Start with one, all-inclusive cluster
At each step, split a cluster until each cluster contains a point (or
there are k clusters)
Traditional hierarchical algorithms use a similarity or
distance matrix
Agglomerative Clustering Algorithm
More popular hierarchical clustering technique
Basic algorithm is straightforward
1.
Compute the proximity matrix
2.
Let each data point be a cluster
3.
Repeat
.
erge e wo c oses c us ers
5.
Update the proximity matrix
6.
Until
only a single cluster remains
Key operation is the computation of the proximity of
two clusters
–
Different approaches to defining the distance between
Starting Situation
Start with clusters of individual points and a
proximity matrix
p1 p3 p2 p1 p2 p3 p4 p5. . .
p5 p4 . . .Proximity Matrix
Intermediate Situation
After some merging steps, we have some clusters
C3 C2 C1 C1 C3 C4 C2 C3 C4 C5 C1 C4 C2 C5 C5
Proximity Matrix
Intermediate Situation
We want to merge the two closest clusters (C2 and C5) and
update the proximity matrix.
C3 C2 C1 C1 C3 C4 C2 C3 C4 C5 C1 C4 C2 C5 C5
Proximity Matrix
After Merging
The question is “How do we update the proximity matrix?”
C3 ? ? ? ? ? ? C2
U
C5 C1 C1 C3 C2U
C5 C3 C4 C1 C4 C2U
C5 ? C4Proximity Matrix
How to Define Inter-Cluster Similarity
p1 p3 p5 p4 p2 p1 p2 p3 p4 p5. . .
Similarity?
.
.
.
MAX
Group Average
Distance Between Centroids
Other methods driven by an objective
function
– Ward’s Method uses squared error
How to Define Inter-Cluster Similarity
p1 p3 p5 p4 p2 p1 p2 p3 p4 p5. . .
.
.
.
Proximity Matrix
MAX
Group Average
Distance Between Centroids
Other methods driven by an objective
function
How to Define Inter-Cluster Similarity
p1 p3 p5 p4 p2 p1 p2 p3 p4 p5. . .
.
.
.
Proximity Matrix
MAX
Group Average
Distance Between Centroids
Other methods driven by an objective
function
How to Define Inter-Cluster Similarity
p1 p3 p5 p4 p2 p1 p2 p3 p4 p5. . .
.
.
.
Proximity Matrix
MAX
Group Average
Distance Between Centroids
Other methods driven by an objective
function
How to Define Inter-Cluster Similarity
p1 p3 p5 p4 p2 p1 p2 p3 p4 p5. . .
× × × × ××××.
.
.
Proximity Matrix
MAX
Group Average
Distance Between Centroids
Other methods driven by an objective
function
Cluster Similarity: MIN or Single Link
Similarity of two clusters is based on the two
most similar (closest) points in the different
clusters
– Determined by one pair of points, i.e., by one link in
the proximity graph.
I1
I2
I3
I4
I5
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
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 5Hierarchical Clustering: MIN
1
2
5
1
3
5
0.15 0.2Nested
Clusters
Dendrogram
4
6
4
0 3 6 2 5 4 10.05 0.1
Strength of MIN
Original
Points
Two
Clusters
Limitations of MIN
Original Points
Two Clusters
Cluster Similarity: MAX or Complete Linkage
Cluster Similarity: MAX or Complete Linkage
Similarity of two clusters is based on the two least
Similarity of two clusters is based on the two least
similar (most distant) points in the different
similar (most distant) points in the different
clusters
clusters
–
–
Determined by all pairs of points in
Determined by all pairs of points in the two clusters
the two clusters
II1
1
II2
2
II3
3
II4
4
II5
5
I1
I1 1
1.0
.00
0 0
0.9
.90
0 0
0.1
.10
0 0
0.6
.65
5 0
0.2
.20
0
I2
I2 0
0.9
.90
0 1
1.0
.00
0 0
0.7
.70
0 0
0.6
.60
0 0
0.5
.50
0
I3
I3 0
0.1
.10
0 0
0.7
.70
0 1
1.0
.00
0 0
0.4
.40
0 0
0.3
.30
0
I4
I4 0
0.6
.65
5 0
0.6
.60
0 0
0.4
.40
0 1
1.0
.00
0 0
0.8
.80
0
I5
I5 0
0.2
.20
0 0
0.5
.50
0 0
0.3
.30
0 0
0.8
.80
0 1
1.0
.00
0
11 22 33 44 55Hierarchical Clustering: MAX
Hierarchical Clustering: MAX
0.25 0.25 0.3 0.3 0.35 0.35 0.4 0.4
1
1
2
2
5
5
2
2
5
5
4
4
N
Ne
es
stte
ed
d
C
Cllu
us
stte
errs
s
D
De
en
nd
drro
og
grra
am
m
3 6 4 1 2 5 3 6 4 1 2 5 00 0.05 0.05 0.1 0.1 0.15 0.15 0.2 0.2
4
4
6
6
1
1
3
3
Strength of MAX
Strength of MAX
O
Orriig
giin
na
al
l P
Po
oiin
ntts
s
T
Tw
wo
o C
Cllu
us
stte
errs
s
Limitations of MAX
Original Points
Two Clusters
•
Tends to break large clusters
Cluster Similarity: Group Average
Cluster Similarity: Group Average
Proximity of two clusters is the average of pairwise proximity
Proximity of two clusters is the average of pairwise proximity
between points in the two clusters.
between points in the two clusters.
Need to use average connectivity for scalability since total
Need to use average connectivity for scalability since total
|
|
|Cluster
|Cluster
|
|
|Cluster
|Cluster
))
p
p
,
,
p
p
proximity(
proximity(
))
Cluster
Cluster
,
,
Cluster
Cluster
proximity(
proximity(
j j ii Cluster Cluster p p ClusterCluster p p j j ii j j ii j j j j ii ii ∗ ∗ = =∑
∑
∈ ∈ ∈ ∈proximity favors large clusters
proximity favors large clusters
II1
1
II2
2
II3
3
II4
4
II5
5
I1
I1 1
1.0
.00
0 0
0.9
.90
0 0
0.1
.10
0 0
0.6
.65
5 0
0.2
.20
0
I2
I2 0
0.9
.90
0 1
1.0
.00
0 0
0.7
.70
0 0
0.6
.60
0 0
0.5
.50
0
I3
I3 0
0.1
.10
0 0
0.7
.70
0 1
1.0
.00
0 0
0.4
.40
0 0
0.3
.30
0
I4
I4 0
0.6
.65
5 0
0.6
.60
0 0
0.4
.40
0 1
1.0
.00
0 0
0.8
.80
0
I5
I5 0
0.2
.20
0 0
0.5
.50
0 0
0.3
.30
0 0
0.8
.80
0 1
1.0
.00
0
11 22 33 44 55Hierarchical Clustering: Group Average
Hierarchical Clustering: Group Average
0.15 0.15 0.2 0.2 0.25 0.25 1 1 2 2 5 5
2
2
5
5
4
4
N
Ne
es
stte
ed
d
C
Cllu
us
stte
errs
s
D
De
en
nd
drro
og
grra
am
m
3 6 4 1 2 5 3 6 4 1 2 5 00 0.05 0.05 0.1 0.1 3 3 4 4 6 6
1
1
3
3
Hierarchical Clustering: Group Average
Hierarchical Clustering: Group Average
Compromise between Single and Complete
Compromise between Single and Complete
Link
Link
Strengths
Strengths
–
–
Less susceptible to noise and outliers
Less susceptible to noise and outliers
Limitations
Limitations
–
Cluster Similarity: Ward’s Method
Similarity of two clusters is based on the increase
in squared error when two clusters are merged
– Similar to group average if distance between points is
distance squared
ess suscept e to no se an out ers
Biased towards globular clusters
Hierarchical analogue of K-means
Hierarchical Clustering: Comparison
MIN
MAX
1 2 3 4 5 61
2
5
3
4
1 2 3 4 5 61
2
3
4
5
Group Average
Ward’s Method
1 2 3 4 5 61
2
5
3
4
1 2 3 4 5 61
2
5
3
4
Hierarchical Clustering: Time and Space requirements
O(N
2
) space since it uses the proximity matrix.
– N is the number of points.
O(N
3
) time in many cases
– There are N ste s and at each ste the size N
2
proximity matrix must be updated and searched
– Complexity can be reduced to O(N
2
log(N) ) time for
Hierarchical Clustering: Problems and Limitations
Once a decision is made to combine two 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
MST: Divisive Hierarchical Clustering
Build MST (Minimum Spanning Tree)
– Start with a tree that consists of any point
– 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
MST: Divisive Hierarchical Clustering
DBSCAN
DBSCAN is a density-based algorithm.
–
Density = number of points within a specified radius (Eps)
–
A point is a core point if it has more than a specified number
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
DBSCAN Algorithm
Eliminate noise points
DBSCAN: Core, Border and Noise Points
Original Points
Point types: core,
border and noise
Eps = 10, MinPts = 4
When DBSCAN Works Well
Original Points
Clusters
•
Resistant to Noise
When DBSCAN Does NOT Work Well
Original Points
(MinPts=4, Eps=9.75).
•
Varying densities
DBSCAN: Determining EPS and MinPts
Idea is that for points in a cluster, their k
th
nearest
neighbors are at roughly the same distance
Noise points have the k
th
nearest neighbor at farther
distance
So, plot sorted distance of every point to its k
th
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?
But “clusters are in the eye of the beholder”!
Then why do we want to evaluate them?
– To avoid finding patterns in noise
– To compare clustering algorithms
– To compare two sets of clusters
– To compare two clusters
Clusters found in Random Data
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 y Random Points 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 y DBSCAN 0 0.2 0.4 0.6 0.8 1 x 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 y K-means 0 0.2 0.4 0.6 0.8 1 x 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 y Complete Link1.
Determining the clustering tendency of a set of data, i.e.,
distinguishing whether non-random structure actually exists in the
data.
2.
Comparing the results of a cluster analysis to externally known
results, e.g., to externally given class labels.
3.
Evaluating how well the results of a cluster analysis fit the data
without
reference to external information.
Different Aspects of Cluster Validation
- Use only the data
4.
Comparing the results of two different sets of cluster analyses to
determine which is better.
5.
Determining the ‘correct’ number of clusters.
For 2, 3, and 4, we can further distinguish whether we want to
evaluate the entire clustering or just individual clusters.
Numerical measures that are applied to judge various aspects
of cluster validity, are classified into the following three types.
– External Index: Used to measure the extent to which cluster labels
match externally supplied class labels.
Entropy
– Internal Index: Used to measure the goodness of a clustering
structure
without
respect to external information.
Measures of Cluster Validity
Sum of Squared Error (SSE)
– Relative Index: Used to compare two different clusterings or
clusters.
Often an external or internal index is used for this function, e.g., SSE or
entropy
Sometimes these are referred to as criteria instead of indices
– However, sometimes criterion is the general strategy and index is the
numerical measure that implements the criterion.
Two matrices
–
Proximity Matrix
–
“Incidence” Matrix
One row and one column for each data point
An entry is 1 if the associated pair of points belong to the same cluster
An entry is 0 if the associated pair of points belongs to different clusters
Measuring Cluster Validity Via Correlation
–
Since the matrices are symmetric, only the correlation between
n(n-1) / 2 entries needs to be calculated.
High correlation indicates that points that belong to the
same cluster are close to each other.
Not a good measure for some density or contiguity based
Measuring Cluster Validity Via Correlation
Correlation of incidence and proximity matrices
for the K-means clusterings of the following two
data sets.
0.8 0.9 1 0.8 0.9 1 0 0.2 0.4 0.6 0.8 1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 x y 0 0.2 0.4 0.6 0.8 1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 x y Corr = -0.9235 Corr = -0.5810
Order the similarity matrix with respect to cluster
labels and inspect visually.
Using Similarity Matrix for Cluster Validation
0.7 0.8 0.9 1 10 20 30 0.7 0.8 0.9 1 0 0.2 0.4 0.6 0.8 1 0 0.1 0.2 0.3 0.4 0.5 0.6 x y Points P o i n t s 20 40 60 80 100 40 50 60 70 80 90 100 Similarity0 0.1 0.2 0.3 0.4 0.5 0.6
Using Similarity Matrix for Cluster Validation
Clusters in random data are not so crisp
i n t s 10 20 30 40 50 0.5 0.6 0.7 0.8 0.9 1 0.5 0.6 0.7 0.8 0.9 1 Points P o 20 40 60 80 100 60 70 80 90 100 Similarity0 0.1 0.2 0.3 0.4 .
DBSCAN
0 0.2 0.4 0.6 0.8 1 0 0.1 0.2 0.3 0.4 . xo i n t s 10 20 30 40 50 0.5 0.6 0.7 0.8 0.9 1
Using Similarity Matrix for Cluster Validation
Clusters in random data are not so crisp
0.5 0.6 0.7 0.8 0.9 1 y Points P 20 40 60 80 100 60 70 80 90 100 Similarity0 0.1 0.2 0.3 0.4
K-means
0 0.2 0.4 0.6 0.8 1 0 0.1 0.2 0.3 0.4 xUsing Similarity Matrix for Cluster Validation
Clusters in random data are not so crisp
0.5 0.6 0.7 0.8 0.9 1 y i n t s 10 20 30 40 50 0.5 0.6 0.7 0.8 0.9 1 0 0.2 0.4 0.6 0.8 1 0 0.1 0.2 0.3 0.4 . x Points P o 20 40 60 80 100 60 70 80 90 100 Similarity0 0.1 0.2 0.3 0.4 .
Complete Link
Using Similarity Matrix for Cluster Validation
1 2 3 6 4 0.4 0.5 0.6 0.7 0.8 0.9 1 500 1000 1500 5 7DBSCAN
0 0.1 0.2 0.3 500 1000 1500 2000 2500 3000 2000 2500 3000
Clusters in more complicated figures aren’t well separated
Internal Index: Used to measure the goodness of a clustering
structure without respect to external information
– SSE
SSE is good for comparing two clusterings or two clusters
(average SSE).
Internal Measures: SSE
Can also be used to estimate the number of clusters
0 1 2 3 4 5 6 7 8 9 10 S S E -6 -4 -2 0 2 4 6
Internal Measures: SSE
SSE curve for a more complicated data set
1 2 3 6 5 4 7
Need a framework to interpret any measure.
–
For example, if our measure of evaluation has the value, 10, is that
good, fair, or poor?
Statistics provide a framework for cluster validity
–
The more “atypical” a clustering result is, the more likely it represents
valid structure in the data
–
Can com are the values of an index that result from random data or
Framework for Cluster Validity
clusterings to those of a clustering result.
If the value of the index is unlikely, then the cluster results are valid
–
These approaches are more complicated and harder to understand.
For comparing the results of two different sets of cluster
analyses, a framework is less necessary.
–
However, there is the question of whether the difference between two
index values is significant
Example
– Compare SSE of 0.005 against three clusters in random data
– Histogram shows SSE of three clusters in 500 sets of random data
points of size 100 distributed over the range 0.2 – 0.8 for x and y
values
Statistical Framework for SSE
50 1 0.016 0.0180 0.02 0.022 0.024 0.026 0.028 0.03 0.032 0.034 5 10 15 20 25 30 35 40 45 C o u n t 0 0.2 0.4 0.6 0.8 1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 x y
Correlation of incidence and proximity matrices for the
K-means clusterings of the following two data sets.
Statistical Framework for Correlation
0.7 0.8 0.9 1 0.7 0.8 0.9 1 0 0.2 0.4 0.6 0.8 1 0 0.1 0.2 0.3 0.4 0.5 0.6 x y 0 0.2 0.4 0.6 0.8 1 0 0.1 0.2 0.3 0.4 0.5 0.6 x y Corr = -0.9235 Corr = -0.5810
Cluster Cohesion: Measures how closely related
are objects in a cluster
– Example: SSE
Cluster Separation: Measure how distinct or
well-separated a cluster is from other clusters
Internal Measures: Cohesion and Separation
– Cohesion is measured by the within cluster sum of squares (SSE)
– Separation is measured by the between cluster sum of squares
– Where |C
| is the size of cluster i
∑ ∑
∈−
=
i
x
C
i
im
x
WSS
(
)
2
∑
−
=
i i im
m
C
BSS
(
)
2
Internal Measures: Cohesion and Separation
Example: SSE
– BSS + WSS = constant
1
×
×
×
×
m
12
×
×
×
×
3
4
m
×
×
×
×
25
m
10
9
1
9
)
3
5
.
4
(
2
)
5
.
1
3
(
2
1
)
5
.
4
5
(
)
5
.
4
4
(
)
5
.
1
2
(
)
5
.
1
1
(
2 2 2 2 2 2 = + = = − × + − × = = − + − + − + − =Total
BSS
WSS
K=2 clusters:
10
0
10
0
)
3
3
(
4
10
)
3
5
(
)
3
4
(
)
3
2
(
)
3
1
(
2 2 2 2 2 = + = = − × = = − + − + − + − =Total
BSS
WSS
K=1 cluster:
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.
Internal Measures: Cohesion and Separation
Silhouette Coefficient combine ideas of both cohesion and separation,
but for individual points, as well as clusters and clusterings
For an individual point,
i
– Calculate
a= average distance of
i
to the points in its cluster
– Calculate
b= min (average distance of
i
to points in another cluster)
– The silhouette coefficient for a point is then given by
Internal Measures: Silhouette Coefficient
s = – a
a < ,
or s = a - 1
a
≥, not t e usua case
– Typically between 0 and 1.
– The closer to 1 the better.
Can calculate the Average Silhouette width for a cluster or a
clustering
a