Updated Classification Methods using Unlabeled Data

Share:

Description

This package contains a collection of functions which implement data classification. It creates updated classification rules by making use of unlabeled data when obtaining parameter estimates of models. The functions can be implemented over a number of models with the best model selected and displayed.

Details

Package: upclass
Type: Package
Version: 2.0
Date: 2013-11-26
License: GPL-2
LazyLoad: yes

The function upclassifymodel takes an updated approach to typical classification rules on unlabelled data. It obtains initial parameter estimates and membership probabilities using the labeled data only and then iterates through the EM algorithm using the complete data with continuous updating of estimates and probabilities. The example below shows graphically the goodness of fit of such a model using this updated approach and a typical classification method, upclassify. The function upclassify implements upclassifymodel over a desired list of models fitted to the data. The model that best fits the data is returned. For a complete list of function, use library(help="upclass").

Author(s)

Niamh Russell, Laura Cribbin, Thomas Brendan Murphy

Maintainer: Niamh Russell <niamh.russell.1@ucdconnect.ie>

References

Cribbin, L. (2008) upclass: R Package for Performing Updated Classification Rules, unpublished thesis (M.Sc.), University College Dublin.

Examples

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
data(iris)
X<- as.matrix(iris[,-5])
cl<-as.matrix(iris[,5])

indtrain <- sort(sample(1:150, 30))
Xtrain <- X[indtrain,]
cltrain <- cl[indtrain]

indtest <- setdiff(1:150, indtrain)
Xtest <- X[indtest,]
cltest <- cl[indtest]

fitupmodels <- upclassify(Xtrain, cltrain,
     Xtest, cltest)       #testing every model.
plot(fitupmodels)