FactoClass (version 0.7.7)

ward.cluster: Hierarchic Classification by Ward's Method

Description

Performs the classification by Ward's method from the matrix of Euclidean distances.

Usage

ward.cluster(dista, peso = NULL , plots = TRUE, h.clust = 2, n.indi = 25 )

Arguments

dista

matrix of Euclidean distances ( class(dista)=="dist" ).

peso

(Optional) weight of the individuals, by default equal weights

plots

it makes dendrogram and histogram of the Ward's method

h.clust

if it is '0' returns a object of class hclust and a table of level indices, if it is '1' returns a object of class hclust, if it is '2' returns a table of level indices.

n.indi

number of indices to draw in the histogram (default 25).

Value

• It returns an object of class hclust and a table of level indices (depending of h.clust). If plots = TRUE it draws the indices of level and the dendrogram.

Details

It is an entrance to the function h.clus to obtain the results of the procedure presented in Lebart et al. (1995). Initially the matrix of distances of Ward of the elements to classify is calculated: The Ward's distance between two elements to classify $i$ and $l$ is given by: $$W(i,l) = (m_i * m_l)/(m_i + m_i) * dist(i,l)^2$$ where $m_i$ y $m_l$ are the weights and $dist(i,l)$ is the Euclidean distance between them.

References

Lebart, L. and Morineau, A. and Piron, M. (1995) Statisitique exploratoire multidimensionnelle, Paris.

Examples

data(ardeche)
ca <- dudi.coa(ardeche$tab,scannf=FALSE,nf=4)

 ward.cluster( dista= dist(ca$li), peso=ca$lw )

 dev.new()
 HW <- ward.cluster( dista= dist(ca$li), peso=ca$lw ,h.clust = 1)
 plot(HW)
 rect.hclust(HW, k=4, border="red")