I have seen a few examples implemting the gbm algorithm on youtube using the titanic dataset. These examples have turned some factor variables into dummy/indicator variables when GBM can handle factor variables by internally creating dummies. I am working on an example with healthcare data and I have ended up transforming some factor variables with less than 10 levels into dummy variables. I want to ask if such a transformation can create a problem when it comes to classification accuracy?
My other questions are:
- What is the benefit of using a dummy variable with gbm compared to using a factor variable with less than 10 levels?
- Does anyone has any literature recommending or presenting a contrast to the use of dummy variables with GBM?
I will appreciate help in this regard.