Master Association Rule Mining: Apriori Algorithm Explained with Python

1. Association Rule Mining Concepts

Association rules describe the dependency between items in a dataset, allowing the prediction of one item based on the presence of others. First proposed by Agrawal, Imielinski, and Swami in 1993, they are widely used in market basket analysis, such as the classic example of diapers and beer purchases.

Diapers & Beer

Milk & Bread

Document recommendations

Book recommendations

Medical treatment combos

Banking cross‑sell opportunities

2. Support and Confidence

For a rule R: A ⇒ B, support is the proportion of transactions containing both A and B, calculated as Support(A⇒B)=count(A∪B)/|D|. Confidence measures the reliability of the rule, computed as Confidence(A⇒B)=Support(A⇒B)/Support(A). Example: if 4 transactions contain orange juice and 2 of them also contain coke, the confidence is 0.5 and the support is 0.4.

3. Strong vs. Weak Rules

A rule is considered strong when its support is greater than or equal to a user‑defined minimum support (supmin) and its confidence meets or exceeds a minimum confidence (confmin); otherwise it is a weak rule.

4. Apriori Algorithm for Frequent Itemsets

The Apriori algorithm discovers frequent itemsets through iterative scanning of the transaction database:

Identify all 1‑item candidates and prune those below the minimum support to obtain L1.

Generate 2‑item candidates from L1, count, and prune to get L2.

Repeat the join‑and‑prune steps to form higher‑order candidates (C3, C4, …) until no new frequent itemsets are found.

Frequent itemsets are those whose occurrence count meets the minimum support threshold; they form the basis for generating strong association rules.

5. Python Implementation Overview

Although no built‑in Scikit‑learn function exists for Apriori, the algorithm can be implemented in Python to compute frequent itemsets and calculate support and confidence for each rule. The implementation follows the same steps: generate candidate itemsets, prune by support, and then evaluate confidence to retain strong rules.