Zero inflated data

Hello Everyone,

I’ve been dealing with data where the target variable has like 80% of zeros (The Target variable is actually the count of customers that click an ad). So, the models is performing sloppy. Can anyone suggest a better approach other than Downsampling or Upsampling the data?

Regards,
Ajay Kalidindi

Maybe using F1, precision, recall, AUC , ROC…for measuring the accuracy of the model instead of just conventional accuracy it might be better for these unbalanced dataset. Check them out.

Please see class_weights as well.