How to work with highly imbalanced dataset (22 of 23 records) for use in machine learning

Question

I have a dataset of customers who provided both a description of their shipment and the commodity class it falls into (there are 23 total commodity classes). In many cases