Search code examples
data-miningrule

Why Confidence does not consider B in Association rule mining

For the ARM rule A->B, confidence can be calculated as AUB/A, which gives you out of all transactions in which A is present, how many have A and B together.

My feeling is that in confidence calculation we are giving more importance to A rather than B. But B should also be considered equally valuable in determining the importance of a rule.

Precisely, I want to say that there should be another term (may be called inverse of confidence) which equals AUB/B.

Any kind of explanations are welcome....

Solution

  • The equation wasn't chosen for fun or intuition.

    There is a reason why the rule is A -> B, and not A <- B...

    The mathematical reason to define it this way is called conditional probabilitiy.

    The decision rule A -> B is a conditional rule. If A then B. Not the other way round... the formula you propose is for if B then A.