# Boosted decision trees using Matlab

I would like to experiment with classification problems using boosted decision trees using Matlab. In the paper An Empirical Comparison of Supervised Learning Algorithms this technique ranked #1 with respect to the metrics the authors proposed. My question is, is there a library in Matlab for this type of supervised classification?

The function fitensemble(...) has multiple techniques which I'm finding difficult to understand. For example, the 'LSBoost' technique is relevant for regression problems while I'm interested in classification only. Moreover, in a similar question posted this exact question was asked and answered with regards to R, not Matlab.

• well, sort of answered. no substantial information was given in the answer to the simailar question regarding calibrating predictions Commented May 7, 2014 at 2:05
• Its best not to boost decision trees. Regression trees are used in modern boosting. Commented Nov 29, 2016 at 14:04

## 1 Answer

Did you take a look at the documentation for fitensemble? Here it is: http://www.mathworks.com/help/stats/fitensemble.html There is a list of ensemble algorithms for classification and regression close to the top of the page. An example at the bottom shows how to grow an ensemble of decision trees by AdaBoost.

You can find more examples and explanations on this page http://www.mathworks.com/help/stats/ensemble-methods.html

By default trees for boosting are stumps. To see all default settings, click on the templateTree link in the Learners section of the fitensemble doc page. MinLeaf and MinParent are the two parameters that control the tree size. The doc for MinParent says: For boosting, the default is the number of training observations.

• If you are referring to the example with AdaboostM1, I believe the 'Tree' option is actually a stump, i.e. a tree with three nodes (1 root and 2 terminal). This option is not well documented in my opinion. Commented May 7, 2014 at 8:41
• That's correct - by default trees for boosting are stumps. To see all default settings, click on the templateTree link in the Learners section of that doc page. MinLeaf and MinParent are the two parameters that control the tree size. The doc for MinParent says: For boosting, the default is the number of training observations. Commented May 7, 2014 at 12:26
• You can find more examples and explanations on this page mathworks.com/help/stats/ensemble-methods.html#bsvjyi5 The Prepare the Weak Learners section would help. Commented May 7, 2014 at 12:31
• That's what I was looking for. If you can kindly phrase your comments into an answer to my question, I'll mark it as the answer for others to see. Commented May 7, 2014 at 13:23