2

I used randomForest( ) function in R. But R can't deal variables that more 32factor levels. But my factor variables are very important thing. So I want to use this variable. Then, how to deal this variable?

I know why randomForest( ) can't handle factor variable that have more 32 levels. But I'm wondering that how to use this categorical variable.

Binning the categorical variable??? or just delete?? ??

kjetil b halvorsen
  • 77,844
이순우
  • 51
  • Here is a similar question: https://stats.stackexchange.com/questions/270201/pooling-levels-of-categorical-variables-for-regression-trees/275867#275867 – kjetil b halvorsen May 19 '17 at 10:18

2 Answers2

2

You can either try what Sainath said, or you can do one hot encoding of your variable, which will make a different column for each of your factor levels and the values in those columns would be 0 or 1 depending on whether that row contained that factor level. You can read how to do it here. I personally suggest doing one hot encoding of every factor variable regardless of whether it has more than 32 levels or not.

Dhruv Mahajan
  • 227
0

One way would be to combine the levels with low frequency into a single level. This will reduce the total number of levels, and may even bring it to <32 levels.

Another way would be to calculate the frequency of each level in the variable, and use this as a feature.

Yet another way is the Impact coding approach. For more details regarding the Impact coding technique, see this article http://www.win-vector.com/blog/2012/07/modeling-trick-impact-coding-of-categorical-variables-with-many-levels/.

Sainath Adapa
  • 5