Fast Frequent Itemset Mining using Compressed Data Representation

1
Fast Frequent Itemset Mining using Compressed
Data Representation

• Raj P. Gopalan Yudho Giri Sucahyo
• School of Computing
• Curtin University of Technology
• Bentley, Western Australia 6102
• raj, sucahyoy_at_computing.edu.au

2
Outline

• Introduction
• Data Structure
• Running Examples
• Performance Study
• Comparison with other algorithms
• Further Work
• Conclusion

3
Introduction

• Association rule mining
• Finds interesting patterns or relationships among
  items in a given data set.
• Two steps
• Find the frequent itemsets.
• Use the result of Step 1 to generate association
  rules.
• Step 1 computationally very expensive.
• So, focus of significant research effort.

4
Finding Frequent Itemsets

• Two general approaches
• Candidate generation-and-test
• Apriori (Agrawal et al., SIGMOD93) and its
  variants
• Pattern Growth Approach
• FP-Growth (Han et al., SIGMOD00)
  H-Mine (Pei et al., ICDM01)
• Data Structures
• Array based H-struct in H-Mine
• Tree based FP-Tree in FP-Growth

5
Contributions

• Propose a new algorithm (named CT-Mine) for
  mining complete frequent itemsets directly from
  the compressed prefix tree.
• Compare the performance with Apriori, Eclat (Zaki
  2000), FP-Growth algorithms.

6
Association Rules

• Given a database of transactions containing
  various items, statements of the form
• A ? B (10, 80)
• 80 of transactions that purchase A also purchase
  B and 10 of all transactions contain both of
  them.

7
Transaction Tree
8
CT-Mine Data Structure

• ITEMTABLE
• Every item, with its support and a pointer to
  the root of the subtree of the item.

• COMPRESSED TRANSACTION TREE
• All transactions of the database containing
  frequent items. Only frequent items will be
  stored in the tree.

9
CT-Mine Algorithm

• Three steps
• Identify Frequent Items and Initialize ItemTable
• Construct the Compressed Transaction Tree
• Mining
• Algorithm details in paper.

10
Example
11
Example

• 5 (5), 45 (4), 345 (3), 2345 (1), 12345 (0),

12
Performance Study on Connect-4

• Connect-4
• 67,557 trans
• of items 129
• Avg
• 43 items/trans

13
Performance Study on Pumsb

• Pumsb
• 49,046 trans
• of items 2,087
• Avg
• 50 items/trans

14
Performance Study on Chess

• CHESS
• 3,196 trans
• - of items 75
• Avg
• 37 items/trans

15
Comparing with Apriori

• Apriori 4.04
• also uses a prefix tree
• suffers from poor performance since it has to
  traverse the database many times to test the
  support of candidate itemsets.
• CT-Mine
• follows the pattern growth approach

16
Comparing with Eclat

• Eclat
• Vertical representation
• Uses tid-intersection in its mining process.
• There is no compression scheme in Eclat.

17
Comparing with FP-Growth

• FP-Growth uses FP-Tree.
• FP-Tree is used to group transactions.
• No compression scheme to combine identical
  subtrees as in CT-Mine.
• Number of nodes in CT-Mine will be less by up to
  half of the nodes.
• CT-Mine does not need to build conditional
  FP-Tree.

18
Further Work

• Extend CT-Mine for very large databases.
• Integrate constraints into CT-Mine.

19
Conclusion

• A new algorithm (CT-Mine) for efficient discovery
  of frequent itemsets has presented.
• A compressed prefix tree is used for efficient
  memory usage.
• The performance of CT-Mine against Apriori, Eclat
  and FP-Growth on various data sets is presented.
  The results show that CT-Mine outperformed all
  others at most of the common support levels used
  in mining.

20
Thank you !!
http//www.cs.curtin.edu.au/sucahyoy