📊 Association Rule Learning: Apriori, ECLAT, and FP-Growth Explained

Have you ever wondered how Amazon knows that people who buy a phone also buy a phone case? This magic comes from Association Rule Learning, a machine learning technique used to discover interesting relationships between variables in large datasets.

Let’s break it down in a way that even a non-technical reader can understand.

🤝 What Is Association Rule Learning?

Association Rule Learning is about finding hidden patterns or relationships in data — especially useful in market basket analysis, where retailers want to know what products are often bought together.

🔑 Key Terms You Should Know

Before diving into the algorithms, here are three important terms:

Term Meaning Support How frequently an item or itemset appears in the dataset. Confidence How often the rule is true (e.g., if A then B). Lift How much more likely B is to be bought when A is bought.

🧠 Example Dataset

Imagine a small grocery store dataset of 5 transactions:

Transaction Items Bought T1 Milk, Bread, Butter T2 Bread, Butter T3 Milk, Bread T4 Milk, Butter T5 Bread, Butter, Jam

Now we want to find patterns like:

Get Adarshajay’s stories in your inbox

Join Medium for free to get updates from this writer.

Remember me for faster sign in

“If someone buys Bread and Butter, they are also likely to buy Jam.”

🧮 1. Apriori Algorithm

💡 How It Works:

• Starts with individual items
• Combines them to form pairs, triplets, etc.
• Eliminates combinations that don’t meet a minimum support
• Repeats this process

✅ Example:

1. Count how often each item appears (Support)
2. Filter out items below minimum support (say, 60%)
3. Combine frequent items to form larger sets
4. Generate rules like:

{Bread, Butter} → {Jam}
Support: 20% | Confidence: 50%

🔍 Pros:

• Easy to understand
• Good for small datasets

⚠️ Cons:

• Scans the database many times
• Slower on large data

🌲 2. ECLAT Algorithm

💡 How It Works:

• Uses Vertical Layout: keeps track of which transactions contain each item (TID list)
• Finds frequent itemsets using set intersections

✅ Example:

• Milk: {T1, T3, T4}
• Bread: {T1, T2, T3, T5}
• Milk ∩ Bread = {T1, T3} → Support = 2

{Milk, Bread} is frequent if support ≥ 2

🔍 Pros:

• Faster than Apriori
• Reduces number of scans

⚠️ Cons:

• Uses more memory (stores all transaction IDs)

⚡ 3. FP-Growth Algorithm

💡 How It Works:

• Builds a compact tree structure called FP-Tree
• Avoids generating all possible item combinations
• Recursively mines frequent patterns from the tree

✅ Example:

It compresses the dataset into a tree like this:

Milk → Bread → Butter
Bread → Butter
Milk → Bread
Milk → Butter
Bread → Butter → Jam

Then, it efficiently finds patterns from this structure.

{Bread, Butter} → {Jam} is discovered without scanning the dataset repeatedly.

🔍 Pros:

• Very fast and scalable
• Efficient with large datasets

⚠️ Cons:

• FP-Tree can be hard to build and manage in memory

🔄 Summary Comparison

Algorithm DB Scans Memory Use Speed Best Use Case Apriori Multiple Low Slow Small datasets ECLAT Few Medium Medium Dense datasets FP-Growth 1–2 High Very Fast Large-scale pattern mining

💡 Real-Life Applications

• 🛒 Retail: Suggesting products to add to a shopping cart
• 🎶 Streaming: Recommending songs based on listening history
• 🏥 Healthcare: Finding co-occurring symptoms or treatments
• 📚 E-learning: Suggesting courses based on learning paths

🧾 Conclusion

Association Rule Learning helps us uncover patterns in data that are not obvious. Each algorithm has its strengths:

Implementation of the algorithm — https://github.com/achuajays/Machine-Learning/blob/main/Step_by_Step_Implementation_using_mlxtend.ipynb

• Use Apriori when learning or working with small datasets.
• Use ECLAT if you want better speed and handle memory well.
• Use FP-Growth for performance on big data.