DMCT-NOTES/unit 3/03_FP_Growth_Algorithm.md

FP-Growth Algorithm

FP-Growth (Frequent Pattern Growth) is a faster and more efficient algorithm than Apriori.

Why is it better?

• Apriori scans the database many times (once for size 1, once for size 2, etc.). This is slow.
• FP-Growth scans the database only twice.
  1. First scan: Count frequencies.
  2. Second scan: Build the FP-Tree.

The FP-Tree

An FP-Tree (Frequent Pattern Tree) is a compressed tree structure that stores all the transaction information.

• More frequent items are near the root (top).
• Less frequent items are leaves (bottom).

Steps

1. Count Frequencies: Find support for all items. Drop infrequent ones.
2. Sort: Sort items in each transaction by frequency (Highest to Lowest).
3. Build Tree: Insert transactions into the tree. Shared items share the same path.
4. Mine Tree:
   • Start from the bottom (least frequent item).
   • Build a "Conditional Pattern Base" (all paths leading to that item).
   • Construct a "Conditional FP-Tree".
   • Recursively find frequent patterns.

Divide and Conquer

FP-Growth uses a Divide and Conquer strategy. It breaks the problem into smaller sub-problems (conditional trees) rather than generating millions of candidate sets like Apriori.