Association Rules in Unsupervised Machine Learning

Association Rule Mining: A Complete Guide with Examples & MCQs

Association Rule Mining is one of the most powerful techniques in data mining and business analytics. It helps organizations discover hidden patterns, relationships, and correlations in large datasets — especially in transactional data.

If you've ever seen:

• "Customers who bought X also bought Y"

• Product bundling suggestions on Amazon

• Supermarket combo offers

You've witnessed Association Rule Mining in action.

1. What is Association Rule Mining?

Association Rule Mining is a data mining technique used to discover interesting relationships (associations) between variables in large datasets.

It answers questions like:

• Which products are frequently purchased together?

• What services are commonly used together?

• What patterns exist in customer behavior?

It is widely used in:

• Retail analytics

• E-commerce recommendation systems

• Healthcare data analysis

• Fraud detection

• Cybersecurity event correlation

2. Key Terminology in Association Rules

2.1 Itemset

A collection of one or more items.

Example:

• {Milk, Bread}

• {Laptop, Mouse}

2.2 Support

Support measures how frequently an itemset appears in the dataset.

Support(A)=Number of transactions containing A/Total number of transactions

Example:
If Milk appears in 4 out of 10 transactions:
Support(Milk) = 4/10 = 0.4

2.3 Confidence

Confidence measures how often items in Y appear in transactions that contain X.

Confidence(X→Y)=Support(X∪Y)/Support(X)

It represents the conditional probability:
P(Y | X)

2.4 Lift

Lift measures how much more likely Y is purchased when X is purchased.

Lift(X→Y)=Confidence(X→Y)/Support(Y)

• Lift > 1 → Positive association

• Lift = 1 → Independent

• Lift < 1 → Negative association

Let's calculate:

• Support(Milk) = 3/5 = 0.6

• Support(Bread) = 4/5 = 0.8

• Support(Milk, Bread) = 2/5 = 0.4

Confidence(Milk → Bread) = 0.4 / 0.6 = 0.67
Lift = 0.67 / 0.8 = 0.84

Since Lift < 1, Milk and Bread are weakly negatively associated in this dataset.

🛒 Imagine a Small Grocery Store

Suppose 100 customers visit a shop today.

We want to understand shopping patterns.

✅ 1. What is SUPPORT? (Very Simple)

👉 Support means: How common is something?

If 40 out of 100 people bought Milk,

Then:

• Milk is bought by 40% of customers.

• So, Support of Milk = 40%

🔹 In simple words:

Support tells us how popular an item (or combination) is.

🧺 Example 1: Single Item

100 customers

• 40 bought Milk

Support (Milk) = 40%

That's it. Just popularity.

🧺 Example 2: Two Items Together

Out of 100 customers:

• 30 bought both Milk and Bread together.

Support (Milk & Bread) = 30%

Meaning: 30% of customers buy them together.

💡 So Support = "How frequently does this happen?"

It does NOT tell us whether one causes the other.
It only tells us how common it is.

🚀 2. What is LIFT? (Very Easy Meaning)

👉 Lift answers this question:

"Are two items really connected, or is it just a coincidence?"

🧠 Think Like This:

If:

• 40% buy Milk

• 50% buy Bread

Even randomly, some people will buy both.

So if we see 30% buying both —
Is that special? Or just normal?

Lift tells us that.

🎯 Lift Meaning in Very Simple Words

Lift checks:

👉 "Do people buy Bread MORE OFTEN when they buy Milk?"

📊 Simple Example

Out of 100 customers:

• 40 buy Milk

• 50 buy Bread

• 30 buy both

Now think logically:

If buying Milk had no effect,
Then about 40% × 50% = 20% should buy both randomly.

But we see 30% buying both.

30% is more than 20%.

That means:

Milk and Bread have a real connection.

So:

👉 Lift is greater than 1
👉 There is a positive relationship

4. Apriori Algorithm

The most famous algorithm for association rule mining is:

Apriori Algorithm

Principle:

If an itemset is frequent, all its subsets must also be frequent.

Steps:

1. Generate frequent 1-itemsets

2. Generate candidate 2-itemsets

3. Prune using minimum support

4. Repeat until no more frequent itemsets

5. Generate rules using confidence threshold

🛒 Example 1: Small Grocery Store

You check 10 customer bills.

Bills:

1. Milk, Bread

2. Milk, Bread

3. Milk

4. Bread

5. Milk, Bread

6. Milk

7. Bread

8. Milk, Eggs

9. Bread

10. Milk, Bread

✅ Step 1: Count Single Items

• Milk → 7 times

• Bread → 7 times

• Eggs → 1 time

Now you decide:

👉 "I only care about items bought at least 3 times."

So:

• Milk ✅

• Bread ✅

• Eggs ❌ (too rare, remove)

✅ Step 2: Make Pairs (Only from Popular Items)

Now only check:

Milk & Bread

How many times?

It appears 4 times.

Since 4 ≥ 3 → keep it.

🚀 Final Result

Apriori says:

Popular items:

• Milk

• Bread

• Milk & Bread together

Eggs was removed early.

✔ Saved time
✔ Ignored weak items
✔ Focused on strong patterns

🍿 Example 2: Movie Theatre

You observe 8 customers.

1. Popcorn, Drink

2. Popcorn

3. Drink

4. Popcorn, Drink

5. Popcorn

6. Drink

7. Popcorn, Drink

8. Nachos

Step 1: Count Single Items

• Popcorn → 5

• Drink → 5

• Nachos → 1

Minimum requirement = 2

So:

• Popcorn ✅

• Drink ✅

• Nachos ❌

Step 2: Check Pair

Popcorn & Drink → 3 times

3 ≥ 2 → Keep

Final Pattern

People often buy:

Popcorn + Drink

So theatre can create:

🎯 Combo Offer

🏥 Example 3: Hospital Case

10 patients:

• 6 have Fever

• 6 have Cough

• 5 have both Fever & Cough

• 1 has Headache only

Minimum requirement = 3

Headache ❌ removed

Now check Fever & Cough

Appears 5 times → Strong pattern

Hospital learns:

👉 Fever + Cough often occur together.

💻 Example 4: Online Shopping

Out of 20 customers:

• 12 buy Laptop

• 10 buy Mouse

• 8 buy Laptop & Mouse

• 2 buy Printer

Minimum support = 5

Printer ❌ removed

Laptop & Mouse → 8 times → strong

E-commerce site creates:

"Buy Laptop + Mouse Combo"

🧠 Why Apriori Is Smart

Instead of checking:

Laptop + Printer + Mouse
Milk + Eggs + Butter

It removes weak items first.

So it never wastes time.

5. Applications of Association Rule Mining

1. Market Basket Analysis

Identify products frequently purchased together.

2. Recommendation Systems

Amazon-style product recommendations.

3. Healthcare

Discover disease-drug relationships.

4. Cybersecurity

Detect unusual event combinations (useful in SIEM analysis).

5. Fraud Detection

Find suspicious transaction patterns.

6. Advantages

• Easy to interpret

• Unsupervised learning method

• Highly useful in retail analytics

• Scalable with algorithms like FP-Growth

7. Limitations

• Can generate too many rules

• High computational cost

• Requires careful threshold tuning

• May produce misleading rules without lift