Steps to reproduce.

1) The client and invoice data are merged using the one-to-many option.
2) Duplicates and missing values removed.
3) Select counter_type = 'ELEC', client_catg = 11, counter_coefficient = 1 and invoice_date of 2019 only.
4) The numbering of the categorical variable is reassigned with a new numbering except for counter_statue and target.
5) Standardization applied to the continuous variables.
6) Apply random undersampling with a factor of 0.06 for non-fraud and 1 for fraud. (2nd dataset only)