On-prem document understanding Confidence scores always 100%

Working on an on-prem AI center and automation suite installation.
We trained a document understanding model with relatively small (50 documents) sample size.
Unfortunately, ml model always returns its predictions with 100% confidence.
Note that this is not a validated document (not verified through action center nor validation station) and has never been in training dataset.

I tested over 20+ different documents.

Current dependencies:

Data extraction scope properties:

An example of the confidence levels:

Note that my pdf file is text based so sometimes i get 100% ocr confidence, sometimes not -which is expected-.

@E_Metin

confidence is fine but is the data being extracted correctly?

also sometimes because of excel formatting you might see as 1 can you check by selecting the cell and checking in the top display or check the value from locals pane in studio when in debug mode

cheers

1 Like

@E_Metin

Most probably it’s excel rounding off the confidence score above 0.5 to 1 that’s 100%

Check the score in debug mode as well before writing it to the excel file.

Thanks but this is not an excel formatting error. An example output of Results Object from immediate window:

@E_Metin

are the extracted values correct?

or arethey wrong?

because ocr confidence can be full as well if its clear

cheers

Some of the extracted values are wrong.

Hey @E_Metin
Confidence rate doesn’t mean the extracted value will be definitely correct. But it is a good sign for your solution. Think to add some cross check or business rules.

Thanks, there are existing regex validations but we are getting false positives which can only eliminated by checking extraction scores.
This problem does not happen on Automation Cloud. Only on On-prem Automation Suite installation. AI center version is 2024.10. We will be raising a ticket to UiPath support.