Cart Algorithm in Matlab with more than binary splits?
2 views (last 30 days)
Hello everyone, I want to create a Decision Tree from non-numerical, categorical data (for example I want to predict whether one should play tennis according to the weather conditions which are given categorical like "hot", "mild" or "cold"). The functions in matlab only seem to be using binary splits (maybe I am also wrong), however, I would wish to also allow for splits into more than just 2 branches per node...So do you know how I can have non-numerical predictor data and then allow for more than just binary splits?
A concrete example would be great!
Ilya on 14 May 2012
The CART algorithm described in the book by Breiman et al. uses binary splits only. The MATLAB implementation is based on this algorithm and allows only binary splits. Multiway splits are allowed in other decision tree algorithms such as C4.5.
There is no evidence that multiway splits produce consistently higher accuracy than binary splits, or the other way around.