Cluster Analysis I: Presidency University
Cluster Analysis I: Presidency University
Presidency University
September,2016
Classication
I One way of learning about the nature is to nd out which class
the object belongs .
Classication
I One way of learning about the nature is to nd out which class
the object belongs .
I One way of learning about the nature is to nd out which class
the object belongs .
Classification
Types of classication
Classification
Supervised
Types of classication
Classification
Supervised Unsupervised
Supervised vs Unsupervised Learning
What is
Chair Duck this?
Supervised vs Unsupervised Learning
I Given only x1 , x2 , ....., xn , we try to infer some underlying
structure (i.e. nd similarities and identifying groups).
I The image shows a blue ocean and two land masses with green
vegetation.
Example
I The image shows a blue ocean and two land masses with green
vegetation.
I The image shows a blue ocean and two land masses with green
vegetation.
Clustering
Hierarchical Non-
Clustering Hierarchical
Clustering
Agglomerative Divisive
Hierarchical Clustering
I Suppose we further
split these clusters.
I If we are to further
increase the number
we will split the
second cluster.
Hierarchical Clustering
This step by step clustering process can be expressed using the
following diagram:
Agglomerative vs divisive
I Until all points are in their own cluster, repeatedly split the
group into two resulting in the biggest dissimilarity
Simple Example
Step 7: {1, 2, 3, 4, 5, 6,
7}
Simple Example
We can simply represent this sequence of clustering assignments in
a tree called dendogram.
What's a dendrogram?
I Dendrogram is a convenient graphic to display a hierarchical
sequence of clustering assignments.
What's a dendrogram?
I Dendrogram is a convenient graphic to display a hierarchical
sequence of clustering assignments.
1
daverage (G , H ) = dij
X
nH nG i ∈G ,j ∈H
Average Linkage
1
daverage (G , H ) = dij
X
nH nG i ∈G ,j ∈H
1
daverage (G , H ) = dij
X
nH nG i ∈G ,j ∈H
Single X X X Chaining
Complete X X X Crowding
Average X × ×
Centroid × × × Simple
Linkages in a nutshell
Single X X X Chaining
Complete X X X Crowding
Average X × ×
Centroid × × × Simple
Single X X X Chaining
Complete X X X Crowding
Average X × ×
Centroid × × × Simple