A robust adaptive clustering analysis method for automatic identification of clusters Presenter : Bo-Sheng Wang Authors : P.Y. Mok*, H.Q. Huang, Y.L. Kwok, J.S. Au PR, 2012 1
Jan 26, 2016
A robust adaptive clustering analysis method for automatic identification of clusters
Presenter : Bo-Sheng Wang Authors : P.Y. Mok*, H.Q. Huang, Y.L. Kwok, J.S. Au
PR, 2012
1
Outlines
• Motivation• Objectives• Methodology• Experiments• Compary• Conclusions• Comments
2
Motivation
• Correct cluster numbers do not guarantee that a data set can be properly partitioned in the desired way.
3
Objectives
• The objective of this paper is to propose a robust and adaptive clustering analysis method.
• 1.Produces reliable clustering results
• 2.Identifies the desired cluster number.
4
Methodology-Fuzzy C-mean(FCM)
5
Methodology-Fuzzy C-mean(Example)
6
Methodology-Fuzzy C-mean(Example)
7
Methodology-Fuzzy C-mean(Example)
8
Methodology-Fuzzy C-mean(Example)
9
Mothodology-RAC-FCM
10
Mothodology-RAC-FCM
11
Mothodology-RAC-FCM
12
Mothodology-RAC-FCM
13
Mothodology-Adaptive implementation
14
Experiments-K-mean
15
KM
Experiments-K-mean+RAC-FCM
16
Mothodology-Application
17
Experiments
• When the distribution of cluster number is not stable enough to give the desired number.
Increasing the upper bound of cluster number can.
18
Experiments
19
Experiments
20
How to verification the proposed k parameter?
Experiments
• This paper use the three widely data sets including the Iris data set, Breast Cancer Wisconsin (Diagnostic) data set and Wine data set.
• Step:1.Verified the distribution stability of the cluster number
2.Compared to different cluster validity index methods.
21
Experiments -Iris Data Set
22
Experiments - Breast Cancer Wisconsin (Diagnostic) data set
23
Experiments - Wine data set
24
Experiments -Compary different Data Set
25
Compary-Comparison with the spectral clustering method
26
RAC-FCM Spectral Clustering Method
WIN
Compary-Comparison with cluster ensembles
27
Conclusions
• This paper proposes method no cluster number is needed to define.
• The method is not only robust but also adaptive.
• The method not only identifies the desired cluster number but also ensures reliable clustering results.
28
Comments
• Advantages–We can obtain optimum Result use this method in
cluster analysis.
• Disadvantage– This method is very take the time because of a
program.
• Applications– Cluster Analysis
29