Algorithms for Association Rule Mining: A General Survey and Comparison

1
Algorithms for Association Rule MiningA General
Survey and Comparison
J. Hipp, U. Güntzer, and G. Nakhaeizadeh SIGKDD
Explorations, June 2000. Volume 2, Issue 1
2
Association Rules
3
Problem Definition

• Challenge immense number of rules

Rule restriction (Quality Measures)
4
Itemset I 1,2,3,4
Search Space

• Lattice For I 1,2,3,4

Frequent
Infrequent
Itemset X is Frequent if supp(X) ? minsupp
5
Algorithm reduce Search Space
6
Algorithm Finding Border
Traverse Search Space

• BFS (Breadth-First Search)
• DFS (Depth-First Search)

7
Border Found Tree
Tree For I 1,2,3,4
8
Systematization
Common Algorithms

• Characterize Each Algorithm by Strategy to
• Traverse Search Space
• Determine Support values

9
Comparison of Algorithms
Performance Behavior

• Dataset
• Transaction size 10
• Frequent Itemsets 4
• Transactions generated 100, 000

10
Observations

• Algorithms Balance out on Basket-like Data

Frequent Itemsets size
small
large

• Counting
• Apriori

• Intersecting
• Partition
• Eclat

11
Summary Conclusion
Classic Association Rule restriction
Use of Support Confidence Thresholds
Search Space Minimization
Algorithmic Association rule Mining
Finding Border
Systematization Performance Analysis

• Similar runtime behavior of Algorithms
• No algorithm is fundamentally superior