rpartScore (version 1.0-2)

rpartScore-package: Classification trees for ordinal responses

Description

This package contains functions that allow the user to build classification trees for ordinal responses within the CART framework. The trees are grown using the Generalized Gini impurity function, where the misclassification costs are given by the absolute or squared differences in scores assigned to the categories of the response. Pruning is based on the total misclassification rate or on the total misclassification cost.

Details

Package:	rpartScore
Type:	Package
Version:	1.0-2
Date:	2022-05-25
License:	GPL (>=2)
LazyLoad:	yes

This package contains functions that allow the user to build classification trees for ordinal responses within the CART framework. It is assumed that a set of numerical scores has been assigned to the ordered categories of the response. Two splitting functions are implemented, both based on the generalized Gini impurity function. They use the absolute and the squared differences in scores, respectively, as misclassification costs. In order to select the optimal tree size, pruning can be performed, using two different measures of prediction performance: the total misclassification rate or the total misclassification cost. This package requires the rpart package. The main function in this package is rpartScore. The use of this function is almost the same as the rpart function. The main difference is the presence of two arguments (split and prune) instead of the method argument. The argument split controls the splitting function used to grow the classification tree, by setting the misclassification costs equal to the absolute ("abs" - default option) or to the squared ("quad") differences in scores. The argument prune allows the user to select the prediction performance measure used to prune the classification tree, and can take two values: "mr" (total misclassification rate) or "mc" (total misclassification cost - default option).

References

Breiman L., Friedman J.H., Olshen R.A., Stone C.J. 1984 Classification and Regression Trees. Wadsworth International.

Galimberti G., Soffritti G., Di Maso M. 2012 Classification Trees for Ordinal Responses in R: The rpartScore Package. Journal of Statistical Software, 47(10), 1-25. 10.18637/jss.v047.i10.

Piccarreta R. 2008 Classication Trees for Ordinal Variables. Computational Statistics, 23, 407-427. 10.1007/s00180-007-0077-5.

See Also

rpart

Examples

# NOT RUN {
data("birthwt",package="MASS")

birthwt$Category.s <- ifelse(birthwt$bwt <= 2500, 3,
 	ifelse(birthwt$bwt <= 3000, 2,
 	ifelse(birthwt$bwt <= 3500, 1, 0)))

T.abs.mc <- rpartScore(Category.s ~ age + lwt + race + smoke +
 	ptl + ht + ui + ftv, data = birthwt)

plotcp(T.abs.mc)

T.abs.mc.pruned<-prune(T.abs.mc,cp=0.02)

plot(T.abs.mc.pruned)

text(T.abs.mc.pruned)

T.quad.mr <- rpartScore(Category.s ~ age + lwt + race + smoke + ptl + ht +
 	ui + ftv, split = "quad", prune = "mr", data = birthwt)


# }