Unlocking Market Insights: A Hands‑On Guide to Association Rule Mining with Apriori

1. Association Rule Mining Concept and Process

Association rules describe the dependency between items; if two or more items are related, the presence of one can predict the other. First proposed by Agrawal, Imielinski and Swami in 1993, association rule mining is a key data‑mining technique for extracting valuable relationships from large datasets.

The classic example is the “beer‑and‑diaper” story: analysis of supermarket basket data revealed that customers buying diapers often also bought beer, leading retailers to place these items together to increase sales.

2. Common Applications

Walmart diaper‑beer association

Supermarket milk‑bread association

Baidu Wenku document recommendation

Taobao book recommendation

Medical treatment combinations

Bank cross‑selling services

3. Confidence and Support

A rule is expressed as “If … then …”. Two important measures evaluate a rule: support and confidence.

Support is the proportion of transactions containing both antecedent A and consequent B: Support(A=>B)=count(A|B)/|D|.

Confidence is the conditional probability of B given A: Confidence(A=>B)=count(A|B)/count(A).

For example, if a customer buys orange juice, the confidence that they also buy coke might be 0.5, while the support of this rule could be 0.4.

4. Minimum Support and Frequent Itemsets

An itemset whose support meets or exceeds a user‑defined threshold (minimum support) is called a frequent itemset. Frequent k‑itemsets are denoted Lk. The Apriori algorithm discovers these frequent itemsets.

5. Apriori Algorithm for Frequent Itemsets

The Apriori algorithm iteratively scans the transaction database to find all itemsets whose support is not lower than the minimum support, then generates strong association rules that satisfy both minimum support and minimum confidence.

Key steps:

Generate candidate 1‑itemsets C1 and count their support; discard those below the threshold to obtain L1.

Join L1 to form candidate 2‑itemsets C2, count, and keep those meeting the threshold as L2.

Repeat joining and pruning to generate C3, L3, etc., until no new candidates are produced.

6. Example: From Frequent Itemsets to Strong Rules

Given transactions of items A, B, C, D, E, with minimum support ≥ 50 % and minimum confidence ≥ 50 %, the process identifies frequent itemsets, computes support and confidence for each possible rule, and retains only strong rules that satisfy both thresholds.

7. Limitations of Apriori

Apriori requires multiple scans of the database; for n‑item frequent sets it may need n passes, causing high I/O load and generating a large number of candidate itemsets.

To address this, Han et al. proposed the FP‑growth algorithm in 2000, which builds a compact FP‑tree to mine frequent patterns without repeated database scans.

8. Python Implementation

The article outlines a Python implementation of association rule mining, including calculation of support and confidence, though the actual code is omitted.