Advance Clustering"

advclust is package for advance clustering. Right now advclust provide several Fuzzy Clustering and Consensus Fuzzy Clustering. This package use Object Oriented Programming (S4 for R). Several Algorithms that provided by this package are: - Fuzzy C-Means (FCM) - Gustafson Kessel (GK) - Gath Geva (GG) - Sum Voting Consensus - Product Voting Consensus - Borda Voting Consensus

Fuzzy Clustering Analysis

Fuzzy C-Means

fuzzy.CM() perform fuzzy c-means analysis. More description of this function (parameter setting, description, and return value) explained via ?fuzzy.CM This algorithm used to get sperichal cluster

library(advclust)
data(iris)
fuzzy.CM(X=iris[,1:4],K = 3,m = 2,RandomNumber = 1234)->cl_CM
show(cl_CM)

Gustafson Kessel

fuzzy.GK() perform Gustafson Kessel analysis. More description of this function (parameter setting, description, and return value) explained via ?fuzzy.GK. This algorithm used to get ellipsodial and sperichal cluster

library(advclust)
data(iris)
fuzzy.GK(X=iris[,1:4],K = 3,m = 2,RandomNumber = 1234)->cl_GK
show(cl_GK)

Gath Geva

fuzzy.GG() perform Gath Geva analysis. More description of this function (parameter setting, description, and return value) explained via ?fuzzy.GG This algorithm used to get hyperellipsodial cluster. Use membership that resulted from fuzzy.CM as initial membership to get best result.

library(advclust)
data(iris)
fuzzy.GG(X=iris[,1:4],K = 3,m = 2,RandomNumber = 1234)->cl_GG
show(cl_GG)

Visualization

For visualization this package provide biplot and radar plot

Biplot

Biplot perform visualization with Principal Component Analysis. Use scale =T when unit of variables on data are different.

biploting(cl_CM, iris[,1:4], scale=T)->biplot

Radar Plot

Radar plot can be used to profilling your cluster result via centroid. Please take attention to axis label. 0 indicates mean of variable, 0.5 indicates mean plus half of standar deviation in realted variable, -0.5 indicates mean minus half of standar deviation in related variable, etc.

radar.plotting(cl_CM, iris[,1:4])->radar

Validation

To get best parameter and know how well your result, use validation index that provided. There are Xie Beni, Partition Coefficient, Modified Partition Coefficient, Classification Entropy, Kwon, Tang, and Separation. See details in ?validation.index

validation.index(cl_GK)

Consensus Clustering

Combine several fuzzy cluster result, eg: FCM, GK, GG, to one result. The purpose of this action is to get best stable or robust result. Actually when you run 3 times, with no specific random number, the first result may be different with second time. And choosing best random number is hard to describe. So the alternative is you can combine your result with consensus algorithm.

Right now, this package provide VOTING method. This method inspired from domination in voting schema. Algorithm that use in this package are "sum", "product", "borda". "product" is sensitive when there is low membership on your result.

c_fuzzycluster(cl_GK,cl_GG,cl_CM)->c_consensus
co.vote(c_consensus,"sum")


Try the advclust package in your browser

Any scripts or data that you put into this service are public.

advclust documentation built on May 30, 2017, 1:42 a.m.