5

I need some help in understanding the difference between regression trees and linear model tree.

Regards

Shahzad

Shahzad
  • 1,999
  • 6
  • 35
  • 44

1 Answers1

9

A linear model tree is a decision tree with a linear functional model in each leaf, whereas in classical regression tree (e.g., CART) it is the sample mean of the response variable for statistical units in each leaf (hence, a constant) that is being considered. Linear model trees can be seen as a a form of locally weighted regression, while regression tree are piecewise-constant regression.

For more information on linear model trees, you can consult

Torgo, L. Functional models for regression tree leaves. In Proceedings of the 14th International Conference on Machine Learning, pp. 385–393. Morgan Kaufmann, 1997.

chl
  • 27,771
  • 5
  • 51
  • 71
  • Thanks alot for the exact answer. Are there any software or package available that I can use to construct linear model trees specially in R. Others will also be helpful. – Shahzad Aug 04 '12 at 23:43
  • 1
    @Shahzad Take a look at the [Cubist](http://cran.r-project.org/web/packages/Cubist/index.html) package. Otherwise, you can try Multivariate Adaptive Regression Splines (MARS), see [ESLII](http://www-stat.stanford.edu/~tibs/ElemStatLearn/), §9.4. – chl Aug 05 '12 at 09:19
  • I did some analysis with both the packages. With earth package, I get the warning that the model is Intercept only model so I cannot plot it with plotmo. Is there any possibility that I can plot the model generated by Cubist package? – Shahzad Sep 17 '12 at 09:26