Frequent pattern mining
Finding what goes together in transactions: itemset, support, association rule, confidence, lift, and the Apriori algorithm with its pruning by the minimum support.
- 25 cards
- 6 topics
- Free
The outline, topic by topic
Each topic is weighted the way the exam weighs it.
- Concepts
- Questions
- Sort and link
- The numbers
- Scenarios
- Understand
A few cards from the deck
Frequent itemset: what is it?
Show the answer
A set of items that appear together in at least a given proportion of the transactions, set by a threshold.
A transaction is a till receipt, a session, a basket. The threshold is called the minimum support: it separates what is frequent from what is rare.
Four items A, B, C and D are all frequent. How many candidate pairs must Apriori count at the next level?
- 6, one per pair of items
- 4, one per item
- 8, twice the number of items
- 12, since A–B and B–A differ
Show the answer
6, one per pair of items
A pair is a set, so unordered: 4 × 3 / 2 = 6 (AB, AC, AD, BC, BD, CD). Counting both directions, which gives 12, doubles every pair.
Before running Apriori, you set the …: itemsets that fall below it are discarded.
Show the answer
Before running Apriori, you set the minimum support: itemsets that fall below it are discarded.
Too low, it lets through a mass of uninteresting itemsets and rules; too high, it drops rare but valuable associations.
Association rule: what is it?
Show the answer
An “if A then B” implication extracted from transactions, where A and B are disjoint sets of items, judged by its support, confidence and lift.
It observes co-occurrence, not a cause: “butter → bread” does not say that butter makes people buy bread.