Will XGBoost pose any problem while dealing with categorical variables with more than 2 levels. For example, occupation variable can have values like doctor, engineer, lawyer, data scientist, farmer e.t.c. If so what would be a better method to use in that case?
1 Answer
$\begingroup$
$\endgroup$
I think you should be more specific about what you mean by "fail". As an example, a practitioner could consider an xgboost
model as a failure if it achieves < 80% accuracy.
Nevertheless, there are some annoying quirks in xgboost
which similar packages don't suffer from:
xgboost
can't handle categorical features whilelightgbm
andcatboost
can.xgboost
can be more memory-hungry thanlightgbm
(although this can be mitigated).xgboost
can be slower thanlightgbm
.