As a rule, the minimum thresholds are defined experimentally. But you should remember about the peculiarities of the thresholds' minimal values:
- If the support value is too great, the algorithm will define meaningless rules at the output. That's why it isn't recommended to set the support to more than 20%.
- If the support value is small, several products may appear within the rules, which are actually random. In this case, you need to watch the lift value.
- If the confidence value is too small, as a rule, this pair cannot be taken into account. The confidence value <7% seems more random.
- The confidence value of more than 85% says that customers practically always buy this pair.