The concept of "Lift" in Sequential Association Rule

[SeanKwon]

Hello,

I did Sequential Association Rules with SAS E-Miner.

Then, I could not find "Lift" in the result.

I realized that "lift" cannot be defined in Sequential Association Rules, but I don't know why.

Is there anyone who explain why the concept of "Lift" cannot be defined in Sequential Association Rules?

[LinusH]

Moved to Mining forum - a more appropriate forum for this question.

[WendyCzika]

This is from the SAS Enterprise Miner Reference Help, hope this answers your question:

Lift is traditionally not calculated for Sequence Analysis. In an Association analysis, the lift is the proportion of time that the right-hand side (RHS) occurs, given that the left-hand-side (LHS) has occurred. This computation is straightforward when there is only one LHS and one RHS, which is the case in Association. The computation becomes much more complex when you have a sequence that has more than one step.

In the sequence,

A -> B -> C -> D

, are you looking at the subset of events where A -> B -> C occurred and then D occurred? Or, are you interested in looking at A -> Band then later C -> D? There is not a unique way to create a lift. Therefore, lift is not calculated.

[SeanKwon]

Thank you for your quick response~!

I'm sorry, but frankly speaking, I cannot understand it well.

I already know that lift is not calculated in sequential association rules.

But, I want to know why.

As you know, in Association Rules, Lift is defined as "Confidence A / P(A)".

For example, in the sequence

            A → B → C → D

I want to know Lift of B→C.

we can calculate Confidence of B→C, and P(B→C).

Thus, I think that we should be able to calculate Lift of B.

Why is not this fomular right?

[WendyCzika]

This paper has some code that might be helpful: http://www2.sas.com/proceedings/forum2007/132-2007.pdf