How best can you display a value which is based on the pattern of other values (not the values themselves)

I have an unusual situation due to some underlying fuzzy logic and machine learning algorithms where I have a priority value of an item, say, low, medium and high.

The priority assessment is based on ~20 underlying calculations and the users are seeking to better understand how the result is arrived at.

Whilst I can (albeit with its own considerations) display the underlying calculation values even from these the user can't easily tell why something has a high, med, low priority as the combination of the underlying calculations isn't easy to rationalise, it's not directly weighted and it's not a simple overall score, average or anything.

The priority is based on how common the pattern of results is, which in turn is based on the data itself rather than anything absolute.

The whole situation isn't aided by the fact the users are more familiar with simpler tools where there are rules which are met or broken and it's easy to understand why a rule has been broken.

A suitable analogy might be how police identify suspicious cars to stop, there are some obvious things like speeding or weaving all over the road, but equally, someone driving too normally might be an indicator they are trying to hide something and the normality could be considered abnormal.

So my longwinded question is whether there is any research into displaying values where the result is related to the underlying data but not directly? or whether there are any generic approaches for aiding understandability when the methodology is necessarily complex.