Location via proxy:   [ UP ]  
[Report a bug]   [Manage cookies]                
Scribd
Upload
770 views

Fuzzy C-Means Clustering - MATLAB FCM

The document describes the fuzzy c-means clustering algorithm and the fcm function in MATLAB. The fcm function performs fuzzy c-means clustering on input data and returns cluster centers and a fuzzy partition matrix. It accepts options to specify the number of clusters, fuzzy partition exponent, maximum iterations, minimum objective function improvement, and whether to display the objective function. Examples demonstrate clustering a data set and configuring the termination conditions.

Uploaded by

Format Seorang Legenda
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
770 views

Fuzzy C-Means Clustering - MATLAB FCM

The document describes the fuzzy c-means clustering algorithm and the fcm function in MATLAB. The fcm function performs fuzzy c-means clustering on input data and returns cluster centers and a fuzzy partition matrix. It accepts options to specify the number of clusters, fuzzy partition exponent, maximum iterations, minimum objective function improvement, and whether to display the objective function. Examples demonstrate clustering a data set and configuring the termination conditions.

Uploaded by

Format Seorang Legenda
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 6

fcm

Fuzzycmeansclustering

collapseallinpage

Syntax
[centers,U]=fcm(data,Nc)

example

[centers,U]=fcm(data,Nc,options)

example

[centers,U,objFunc]=fcm( ___ )

example

Description
[centers,U]=fcm(data,Nc)performsfuzzycmeansclustering
onthegivendataandreturnsNcclustercenters.

example

[centers,U]=fcm(data,Nc,options)specifiesadditional
clusteringoptions.

example

[centers,U,objFunc]=fcm( ___ )alsoreturnstheobjective


functionvaluesateachoptimizationiterationforalloftheprevious
syntaxes.

example

Examples

collapseall

ClusterDataUsingFuzzyCMeansClustering
Loaddata.
loadfcmdata.dat
Find2clustersusingfuzzycmeansclustering.
[centers,U]=fcm(fcmdata,2);
Iterationcount=1,obj.fcn=8.970479
Iterationcount=2,obj.fcn=7.197402
Iterationcount=3,obj.fcn=6.325579
Iterationcount=4,obj.fcn=4.586142
Iterationcount=5,obj.fcn=3.893114
Iterationcount=6,obj.fcn=3.810804
Iterationcount=7,obj.fcn=3.799801
Iterationcount=8,obj.fcn=3.797862
Iterationcount=9,obj.fcn=3.797508
Iterationcount=10,obj.fcn=3.797444
Iterationcount=11,obj.fcn=3.797432
Iterationcount=12,obj.fcn=3.797430
Classifyeachdatapointintotheclusterwiththelargestmembershipvalue.
maxU=max(U);
index1=find(U(1,:)==maxU);
index2=find(U(2,:)==maxU);

OpenThisExample

Plottheclustereddataandclustercenters.
plot(fcmdata(index1,1),fcmdata(index1,2),'ob')
holdon
plot(fcmdata(index2,1),fcmdata(index2,2),'or')
plot(centers(1,1),centers(1,2),'xb','MarkerSize',15,'LineWidth',3)
plot(centers(2,1),centers(2,2),'xr','MarkerSize',15,'LineWidth',3)
holdoff

SpecifyFuzzyOverlapBetweenClusters
Createarandomdataset.

OpenThisExample

data=rand(100,2);
Specifyalargefuzzypartitionmatrixexponenttoincreasetheamountoffuzzyovrelapbetweentheclusters.
options=[3.0NaNNaN0];
Clusterthedata.
[centers,U]=fcm(data,2,options);

ConfigureClusteringTerminationConditions
Loadtheclusteringdata.

OpenThisExample

loadclusterdemo.dat
Settheclusteringterminationconditionssuchthattheoptimizationstopswheneitherofthefollowingoccurs:
Thenumberofiterationsreachesamaximumof25.
Theobjectivefunctionimprovesbylessthan0.001betweentwoconsecutiveiterations.
options=[NaN250.0010];
ThefirstoptionisNaN,whichsetsthefuzzypartitionmatrixexponenttoitsdefaultvalueof2.Settingthe
fourthoptionto0suppressestheobjectivefunctiondisplay.
Clusterthedata.
[centers,U,objFun]=fcm(clusterdemo,3,options);
Viewtheobjectivefunctionvectortodeterminewhichterminationconditionstoppedtheclustering.
objFun
objFun=
54.7257
42.9867
42.8554
42.1857
39.0857
31.6814
28.5736
27.1806
20.7359
15.7147
15.4353
15.4306
15.4305
Theoptimizationstoppedbecausetheobjectivefunctionimprovedbylessthan0.001betweenthefinaltwo
iterations.

RelatedExamples
ClusterQuasiRandomDataUsingFuzzyCMeansClustering
AdjustFuzzyOverlapinFuzzyCMeansClustering

InputArguments

collapseall

dataDatasettobeclustered
matrix
Datasettobeclustered,specifiedasamatrixwithNdrows,whereNdisthenumberofdatapoints.The
numberofcolumnsindataisequaltothedatadimensionality.
NcNumberofclusters

integer
Numberofclusters,specifiedasanintegergreaterthan1.
optionsClusteringoptions
vector
Clusteringoptions,specifiedasavectorwiththefollowingelements:
Option

Description

options(1) ExponentforthefuzzypartitionmatrixU,specifiedasa
scalargreaterthan1.0.Thisoptioncontrolstheamountof
fuzzyoverlapbetweenclusters,withlargervalues
indicatingagreaterdegreeofoverlap.

Default
2.0

Ifyourdatasetiswidewithalotofoverlapbetween
potentialclusters,thenthecalculatedclustercentersmight
beveryclosetoeachother.Inthiscase,eachdatapoint
hasapproximatelythesamedegreeofmembershipinall
clusters.Toimproveyourclusteringresults,decreasethis
value,whichlimitstheamountoffuzzyoverlapduring
clustering.
Foranexampleoffuzzyoverlapadjustment,seeAdjust
FuzzyOverlapinFuzzyCMeansClustering.
options(2) Maximumnumberofiterations,specifiedasapositive
integer.

100

options(3) Minimumimprovementinobjectivefunctionbetweentwo
consecutiveiterations,specifiedasapositivescalar.

1e5

options(4) Informationdisplaytoggleindicatingwhethertodisplay
theobjectivefunctionvalueaftereachiteration,specified
asoneofthefollowing:

0Donotdisplayobjectivefunction.
1Displayobjectivefunction.

IfanyelementofoptionsisNaN,thedefaultvalueforthatoptionisused.
Theclusteringprocessstopswhenthemaximumnumberofiterationsisreachedorwhentheobjective
functionimprovementbetweentwoconsecutiveiterationsislessthanthespecifiedminimum.

OutputArguments

collapseall

centersClustercenters
matrix
Finalclustercenters,returnedasamatrixwithNcrowscontainingthecoordinatesofeachclustercenter.The
numberofcolumnsincentersisequaltothedimensionalityofthedatabeingclustered.
UFuzzypartitionmatrix
matrix
Fuzzypartitionmatrix,returnedasamatrixwithNcrowsandNdcolumns.ElementU(i,j)indicatesthe
degreeofmembershipofthejthdatapointintheithcluster.Foragivendatapoint,thesumofthe

membershipvaluesforallclustersisone.
objFuncObjectivefunctionvalues
vector
Objectivefunctionvaluesforeachiteration,returnedasavector.

MoreAbout

collapseall

Algorithms
Fuzzycmeans(FCM)isaclusteringmethodthatallowseachdatapointtobelongtomultipleclusterswith
varyingdegreesofmembership.
FCMisbasedontheminimizationofthefollowingobjectivefunction
D

Jm =

i =1 j=1

ij xi cj

where
Disthenumberofdatapoints.
Nisthenumberofclusters.
misfuzzypartitionmatrixexponentforcontrollingthedegreeoffuzzyoverlap,withm>1.Fuzzyoverlap
referstohowfuzzytheboundariesbetweenclustersare,thatisthenumberofdatapointsthathave
significantmembershipinmorethanonecluster.
x iistheithdatapoint.
c jisthecenterofthejthcluster.
ijisthedegreeofmembershipofx iinthejthcluster.Foragivendatapoint,x i,thesumofthe
membershipvaluesforallclustersisone.
fcmperformsthefollowingstepsduringclustering:
1.Randomlyinitializetheclustermembershipvalues,ij.
2.Calculatetheclustercenters:
D

cj =

i =1
D

mx
ij i

i =1

m
ij

3.Updateijaccordingtothefollowing:

ij =

1
N
k=1

x c
i j
x c
i k

2
m 1

4.Calculatetheobjectivefunction,J m.
5.Repeatsteps24untilJ mimprovesbylessthanaspecifiedminimumthresholdoruntilafteraspecified
maximumnumberofiterations.
FuzzyClustering

References
[1]Bezdec,J.C.,PatternRecognitionwithFuzzyObjectiveFunctionAlgorithms,PlenumPress,NewYork,1981.

SeeAlso
findcluster|genfis3

IntroducedbeforeR2006a

You might also like