Lessons Learned
This chapter introduces association rules—automatically generated rules about the items most likely to appear together in an order. This is one of the most detailed ways of analyzing transaction information.
Simple one-way association rules specify that when a customer purchases one product (the left-hand side), then the customer is likely to purchase another product (the right-hand side) in the same order. The traditional way of measuring the goodness of these rules is with three measures: support, confidence, and lift. Support measures the proportion of orders where the rule is true. Confidence measures how often the rule is true when it applies. And lift specifies how much better the rule works rather than just guessing.
A better measure, however, is based on the chi-square value introduced in Chapter 3. This gives an indication of how likely it is that the rule is random—and when this likelihood is very small, then the rule is important.
Association rules...