treebagger random forest

In the help file, it is stated that setting Setting 'NVarToSample' argument to any valid value but 'all' invokes Breiman's 'random forest' algorithm.

Is then, the ''OOBVarImp' will be based on GINI index as in the case of Breiman's random forest ?

How to view OOBVarImp information?

Sorry if my questions sounds dumb to you! i am dumb :)

Answer 1

[Ilya]

By setting 'OOBVarImp' to 'on', you fill OOBPermutedVarDeltaError, OOBPermutedVarDeltaMeanMargin, and OOBPermutedVarCountRaiseMargin properties of a TreeBagger object. Do 'help TreeBagger' to see a list of all properties and click on a property for a description. These three do not directly depend on the criterion used to find optimal decision splits (such as Gini). They represent three methods for computing predictor importance proposed by Breiman.

DeltaCritDecisionSplit property is always filled. It is computed by summing changes in the split criterion used to grow trees (such as Gini).

[Sukuchha]

Thanks IIya

Any pointers to features selection based on tree bagger. Tutorial or similiar !

[Ilya]

This doc example has a short section on feature selection.

http://www.mathworks.com/help/toolbox/stats/bsvjye9.html#bsx62vu