Skip to Content

In SAC Smart Discovery simulation how is the base value calculated?

Mar 22 at 09:48 AM


avatar image

What prompted me to ask is that I have a negative which doesn't appear to reflect the data. The target column is 'Number of Non-Conformances' (measure) which can only have a positive or zero value.

All other measures are excluded, so only analysing dimension influencers.


10 |10000 characters needed characters left characters exceeded

is this because there are characteristics which lower the likelihood and which raise the likelihood. If greater weighting of lowering characteristics is selected it is possible to predict below zero?


Hi Angus, please could you add a screenshot or two so that we can picture the use-case and anomaly here?

Kind regards, H


hi Henry thanks hopefully these explain.

negative base value:

then change Booking Status dimension and the expected value is also negative:

also unsure how the change in booking status influence value equates to the change in the overall expected value


1.png (135.9 kB)
2.png (171.5 kB)
* Please Login or Register to Answer, Follow or Comment.

2 Answers

Henry Banks
Apr 05 at 07:22 AM

Hi Angus,

I don't know much about your key influencers here, so difficult to say in context.. what you are seeing here are the various 'contributions by variable' as identified by the automated classification/regression models used in Smart Discovery.

Under the covers, this is the same approach as SAP's Automated Analytics - built using Cloud Platform's "Predictive Business Services " REST APIs. it is moreless this statistical approach:

.. And implemented using this cloud service:

Maybe we can do more to enhance our own documentation, have help tips to guide the user, or at least link out to deeper existing resources? Benton Li Flavia Moser Antoine CHABERT what do you guys think?

.. but yes, SAC doesn't know that the target 'Base Value' can't be negative in your case. Alas, there's no way or setting in SAC to say 'force this continuous variable to show positives-only, or specify a lower limit' before running the algorithm.

A similar problem can be seen when using Time Series Forecast, where negative values are projected. Infact, we're implementing a checkbox 'force positive forecast' to help users only chart positive values, so I'll bring some attention to your requirement.

Infact we're expecting big gains soon with new predictive capabilities (in a number of different directions), so perhaps we can revisit these topics during the UK Partner event in mid-May?

Kind regards,

10 |10000 characters needed characters left characters exceeded
Apr 05 at 09:52 AM

Hi Henry Banks I would agree on your suggestion on a general standpoint. I am not directly in charge of this experience though, so I'll leave the final word on the next steps to Flavia Moser. I think Flavia is out of office for the time being, so hopefully she can get back to us next week at best.

Kind regards,


Show 1 Share
10 |10000 characters needed characters left characters exceeded

Thanks for the input guys. Flavia Moser if it helps explain / investigate, the JF Tech demo also returns negative base values. Although as i say this isn't really the question, it is understandable that a combination of influencers could have a negative result. The question is more how the base value is calculated - ie is it an averaged weighting for all influencers not shown in the simulation - and why does it change when you change simulation values

1.png (134.7 kB)