How would you handle a dataset with millions of rows where 15% of the values in a critical financial column are missing?
