Hello,

From the FAQ section of http://www.sqlserverdatamining.com:

19. What is $AdjustedProbability or PredictAdjustedProbability()?

 The MS Analysis Service DM has a formula to penalize popular items in prediction. Suppose the predicted probability of A and B being purchased are the same, say, 10% based on the customer's basket. Now, suppose A is so popular that 10% of all customers buy while B is not so popular that only 1% of them buy (this is what we call marginal probability that is what we can tell even without looking at that particular customer's basket). In this case, you probably want to recommend B more than A because B would be more interesting to that particular customer. The $AdjustedProbability is a "lift" of the predicted probability over marginal probability. The formula is our internal, undocumented, which may change in the next release, but it looks something like the following:

 AdjustedProbability = PredProb * (1 - MargProb) ^ Constant

As you can see, AdjustedProbability is not a probability per se any more. It's intended to be interpreted as "lift" of probability.

Hope this helps
Shuvro

---

MSFT, SQL Server Data Mining

I'm not aware what exact formula is used to arrive at AdjustedProbability. However, I believe Adjusted Probability is more acceptable in some scenarios.
Because, Adjusted Probability considers marginal probability/probability at overall population level.
For instance, Suppose
probability of predicatable attribute A being "x" turns out to be 50% and A being "y" is 10% for a particular case (this is a predicted probability based on algorithm you applied)
and for a general population, probability of A being "x" in 80% and A being "y" is 1%.
Now, for the above particular case, clearly, A being "y" is more probable (10 times of average probability) whereas A being "x" is lower than average population.. then for this case, Adjusted Probability includes a "lift" making Adjusted Probability for "y" more than 10% while making probability. The simple probability won't consider this factor.
So, depending on the practicality (how practical the adjusted probability turns out to be), you might want to go with AdjustedProbability or the direct probability..
[This is what I understand but not very sure. Also, please double-check in your case if something else is causing it]

---

..hegde

From the FAQ section of http://www.sqlserverdatamining.com:

19. What is $AdjustedProbability or PredictAdjustedProbability()?

 The MS Analysis Service DM has a formula to penalize popular items in prediction. Suppose the predicted probability of A and B being purchased are the same, say, 10% based on the customer's basket. Now, suppose A is so popular that 10% of all customers buy while B is not so popular that only 1% of them buy (this is what we call marginal probability that is what we can tell even without looking at that particular customer's basket). In this case, you probably want to recommend B more than A because B would be more interesting to that particular customer. The $AdjustedProbability is a "lift" of the predicted probability over marginal probability. The formula is our internal, undocumented, which may change in the next release, but it looks something like the following:

 AdjustedProbability = PredProb * (1 - MargProb) ^ Constant

As you can see, AdjustedProbability is not a probability per se any more. It's intended to be interpreted as "lift" of probability.

Hope this helps
Shuvro

---

MSFT, SQL Server Data Mining