1

I have a multiclass classification problem. In the dataset, I have five categorical variables each having 1730, 235, 60,20 and 5 unique categories in each respectively. Apart from that I have 4 numerical features and a target variable having 5 different classes.

I have made sure while collecting the data that classes are well balanced by taking equal number of instances for each class.

As there is problem of high cardinality in the dataset (large number of unique features) I can't use one hot encoding and label encoding.

I have tried target encoding and frequency encoding along with cross validation and grid search. I am still unable to get a good accuracy or f1_score. What should I do?

kjetil b halvorsen
  • 63,378
  • 26
  • 142
  • 467
Shubham Kaushik
  • 41
  • 4
  • 1
    Have a look at https://stats.stackexchange.com/questions/146907/principled-way-of-collapsing-categorical-variables-with-many-levels and other posts tagged [tag:many-categories]. – kjetil b halvorsen Oct 24 '21 at 16:44

0 Answers0