When I am validating data extracted from documents, I’m noticing that when a field is not present on one document, it is sometimes being extracted as as another incorrect value on the page. In the learning instance setup, the field is marked as optional and the confidence is set at 80%. I submit validation feedback by removing the mapping to the value, but each time I process a new document, it extracts the same incorrect data point. I have the same issue with tables that are being extracted for some documents for which the tables are not needed. Even after removing the mappings, the tables are extracted again when processing additional documents. 

Is there a solution for this issue or a best practice for solving this problem?