I have a dataset that consists of a type of unique transaction for all stores in the last two years. Every combination of possibility is captured between stores, unique transaction_Classification, & time_day_id. Many instances exist when a certain transaction does not exist on that day for that unique transaction. I want to build a custom missing value handling that would be built to handle all stores differently based on what unique store type they are.

• Example: New Stores would be treated as similar store types in their particular region. Handle Missing Values for New Stores: (Group By: Store Type, Region, Unique_Transaction_Classification, Month, Week_Day)

• Example: Store_Type: FM stores would be treated as 0’s in their null’s.
  Handle Missing Values for FM stores: (Where Store_Type = ‘FM’) , fillna(0)

I want to replace the null value with a new column that will represent ‘Replace_Null_Transaction_Count_Column ‘

1. If the actual value existed in the original transaction count column
   bring over into new column.
2. All null values will get treated with the custom logic and put in
   this new column.

Example of Data for imputing Missing for Store_Type & Store_Classified

I replaced null’s in the same column with a group by for the store_classified, region, week_day. It replaces new_store & established store with null in the same manner which I do not want. I want to treat different combinations of Store_Classified & Store type with null’s differently until I get all null’s have been handled. I am not sure of the best way to do this step?