1
$\begingroup$

I am trying to predict a multiclass categorical outcome variable by comparing different classifier algorithms.
I've got a dataset that includes two categorical variables that have many labels (>2000). The problem is that the proportion they make up is fairly evenly distributed, which is why I think collapsing the variables wouldn't be a very good option. The following is the distribution for the 10 most frequent labels for the variable:

Government Of Tanzania    20.932829
Danida                     5.186667
Hesawa                     3.833314
World Bank                 2.289588
Rwssp                      2.230204
World Vision               2.186914
Kkkt                       2.135668
Unicef                     1.837318
Tasaf                      1.644216
Dhv                        1.457457
0                          1.362791
Private Individual         1.350987
Dwsp                       1.347113

Since label encoding and one-hot encoding wouldn't be a good option in this case, because of the large number of labels,
what would be a good encoding in this case?

Improve this question
$\endgroup$
1

0

Reset to default

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.