Association rule is an ideally method for collection type problem mining in semiconductor manufacturing. The trick is in setting the support and confidence level.
The initial setting maybe the ((100-baseline yield)/5)/100 for support, 0.7 for confidence.
support(A) = # of event A/ Total ==> If an equipment is malfunction, most of the "bad" must come from that equipment. We regards the low yields as "bad", if more than 1/5 of the bads occurred, it's an candidate.
