Scribd
Upload
14 views

Module-5 Clustering Algorithm

Uploaded by

chaithrak.21.beis
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
14 views

Module-5 Clustering Algorithm

Uploaded by

chaithrak.21.beis
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 31

www.acharya.ac.

in

Hierarchical Clustering Algorithms


Unsupervised Clustering

Click to Edit
Click to Edit
Hierarchical clustering involves creating clusters that have a
predetermined ordering from top to bottom

Click to Edit
Types
• Agglomerative or bottom-up clustering - algorithm
starts by taking all data points as single clusters and
merging them until one cluster is left

Click to Edit
• Divisive method or Top - Down Clustering method

It starts by considering all the data points into a big


single cluster and later on splitting them into smaller
heterogeneous clusters continuously until all data points
are in their own cluster.

Click to Edit
Click to Edit
he distance between two clusters is
the minimum distance between
members of the two clusters
Feature of Cluster Algorithm
• Clusters have equal diameter

• Once cluster formed is irreversible decision.

Single Linkage or MIN Algorithm


Click to Edit
Merge in each step the two clusters whose
two closest members have the smallest
distance
Click to Edit
Example :

Click to Edit
Click to Edit
Step 2 :update the distance matrix for cluster with
minimum distance

Click to Edit
Click to Edit
Click to Edit
Click to Edit
Complete Linkage or MAX or Clique

Click to Edit
Mean shift Clustering Algorithm
(Mode Seeking Algorithm,Sliding Window Algorithm)
● No Need of Prior
Knowledge(non-parametric )

● Slowly moves from initial


position to dense region.
Click to Edit

● Window is called Kernel.

● Window based on Kernel


Density Function.
● Aim - To find underlying data
distribution .
Mean Shift Clustering Algorithm

Click to Edit
Mean shift Clustering Algorithm

Click to Edit
Advantage
• No model Assumptions.
• Suitable for all non-convex shapes.
• Only one parameter of the window(Radius)
• Robust noise

Click to Edit Disadvantage


● Selecting bandwidth is challenging.
● No of clusters cannot be specified.
Mean shift Clustering Algorithm

Click to Edit
K-Means

Click to Edit
Density Based Methods(DBSCAN)
The concept of Density and
connectivity is based on local
distance of neighbours.
Uses to parameter :
1. Size of neighbourhood(eps)
2. Minimum number of points.(m)
Click to Edit 3. Core Point - more than specified no of
points (m) within neighbourhood.
4. Border Point - if has fewer than ‘m’
points but is a neighbour of a core
point.
5. Noise Point -Neither core nor border
point
Density Based Methods(DBSCAN)

Click to Edit
Density Based Methods(DBSCAN)

Click to Edit
Density Based Methods(DBSCAN)

Click to Edit
Grid Based Approach
• Space based approach
used for high dimensional
data

• Partition space into cells.


• Data is fitted on the cells
for cluster formation.
Click to Edit

Three Concepts
● Subspace clustering.
● Concept of dense cells.
● Monotonicity properties
Click to Edit
CLIQUE Algorithm

Click to Edit
Concept of Dense Cell

Click to Edit
THANK YOU
Click to Edit

You might also like