Association Rule Mining
Association rule mining finds if then patterns in transactional data, such as which items tend to be bought together. A rule has the form if a customer buys X then they also buy Y.
Three core metrics
Rules are judged by three numbers.
- Support is the fraction of all transactions that contain both X and Y. It measures how common the pattern is.
- Confidence is the fraction of transactions containing X that also contain Y. It measures how reliable the rule is.
- Lift compares the rule to chance. It is the confidence divided by the overall frequency of Y.
Reading lift
Lift is the key honesty check.
- Lift above one means X and Y appear together more than expected, a genuine association.
- Lift equal to one means they are independent.
- Lift below one means buying X makes Y less likely.
A rule can have high confidence yet a lift near one simply because Y is popular everywhere, so confidence alone can mislead.
The workflow
You first find frequent itemsets whose support clears a threshold, then turn those itemsets into rules that meet a confidence threshold, and finally rank the survivors by lift.
Key idea
Association rule mining surfaces if then item patterns scored by support, confidence, and lift, where lift above one signals a real association.