Explanation of random forest performance difference to when using categories and when using dummy variables

Question

I have some hand coded feature which is a category with values "High", "Low", and "Normal".

I created this feature myself and my problem performance (classification) increased dramatically when using it with expanding these by dummy variables.

Now since I'm trying random forest, I thought I change "High, Low, Normal" to 1, -1, 0 instead.

Now the same model doesn't learn at all.

I thought it should become easier actually for it to split. Does this have to do with me putting normal to 0?

Thank you for any explanation helping me to understand this.

Answer 1

It should work: the variable is ordinal so using numerical values makes sense.

So there's a bug somewhere, here are a few suggestions of things to look at:

• Possibly a type conversion error somewhere: make sure the variable is interpreted as numerical.
• Check whether the model actually uses the variable: if not then it's likely some type error; if yes then I would investigate what goes wrong: for example it might help to plot this variable vs. target in the two cases where the variable is categorical or numerical.
• Maybe some difference between the preprocessing of the training and test set: apply the model on the training set, if the performance is good then it's likely that there's something wrong in the preprocessing of the test set.