Chapter 11 Association Rules

1
Chapter 11Association Rules

• Dongil Kim
• Data Mining Lab.
• March, 31, 2008

2

• From the Survey Result of ??????? (n1,300)
• The most important ability (skill) in the fields
• 1. Excel
• 2. Powerpoint
• 3. Data Analysis
• 4..

3
1. Introduction

• What Goes with What?
• Market Baskets

4
1. Introduction

• What Goes with What?
• Book Stores

Which one seems more associated?
5
2. Association Rules

• Association Rules (Affinity Analysis)
• (A.K.A.) Market Basket Analysis (MBA)
• What goes with what?
• Which items are purchased together?
• Diaper and Beer
• They are purchased together
• Why?

6
2. Association Rules

• What Can We Do?
• Purpose of analyzing market basket
• Wine, Cheese
• Understanding customers
• Understanding merchandise
• Discount Wine (Discount Cheese?)
• Cross-selling
• Display (Short or longer?)

7
2. Association Rules

• Where Can MBA be Applied?
• Retail
• Wine, Cheese, Diaper, Beer, Bread,
  Car-stuffs
• Credit card
• Gas, Family Restaurant, Airlines, Bus, VISA
• Telecom. Company option design
• Bank
• Medical history

8
2. Association Rules

• Association Rule Steps

Define Items and Transactions
Find Rules Support, Confidence, Lift
Analyze Results
9
2. Association Rules

• Define Items (WHAT goes with WHAT)
• POS data
• Items Beer, Whiskey, Coke, 7-up, Nacho,
  Homerun-ball
• Categories Drinks, Beverages, Snacks
• Book Store
• Items Hackers TOEIC, Tomato TOEIC, Hacker TOEFL,
  Norwegian Wood
• Categories TOEIC, Business, Novel

10
2. Association Rules

• Define Items (WHAT goes with WHAT)
• Taxonomy
• Low More actionable result, less confident
  result
• High Less actionable result, more confident
  result
• A trade-off

Merchandise
Drinks, Microwave Instants, Snacks
Coke, Beer, Wine, Milk
Is there any point you can easily find?
Cocacola, Pepsi, Krombacker, Becksdark
Cocacola Zero 500ml, Cocacola 1L
11
2. Association Rules

• Define Transactions (what goes WITH what)
• POS data
• A single market basket
• Ingredient
• A single dish of food vs. A course of the whole
  food
• Bank Account
• Open 3 or 4 different accounts in a single day?
• Credit Card
• Use credit card a day? or for all time?
• Time Space Purpose Character

12
2. Association Rules

• Forms of Data for MBA

Binary n m table (XLMiner, any mathematical
software)
Maybe an original form of transaction data
(XLMiner)
2 1 table (SAS)
13
2. Association Rules

• Statement of the Rule
• IF (condition) THEN (result)
• IF (antecedent) THEN (consequent)
• IF (A) THEN (B)
• A -gt B
• Multi-item conditions
• IF (A,B) THEN (C)
• IF (Radiohead, Oasis) THEN (Suede)
• Reflexive?
• IF (Diaper) THEN (Beer)
• IF (Beer) THEN (Diaper)

14
2. Association Rules

• Support
• How many times the item A appeared in the
  database
• P(A) or P(A and B) or P(AnB)
• Ex)

Support (Beer) 3/6 Support (Wine) 3/6 Support
(Beer, Wine) 1/6 Support (Snack) 5/6
15
2. Association Rules

• Confidence
• Conditional probability (Not reflexive)
• P(result cond) P(cond n result)/P(cond)
• If confidence is large, we can say they have
  association
• Ex)

Conf(Beer -gt Snack) P(Beer n Snack)/P(Beer)
(3/6) / (3/6) 1 Conf(Wine -gt Snack) P(Wine n
Snack)/P(Wine) (2/6) / (3/6) 2/3
16
2. Association Rules

• Lift
• Lift(cond -gt result) Conf (cond -gt
  result)/P(result)
• P(cond n result)/(P(cond)P(result))
• If lift gt 1 they have correlation (we can
  except the result)
• Ex)

Lift(Beer -gt Snack) Conf(Beer -gt
Snack)/P(Snack) 1/(5/6) 6/5 gt1 Lift(Wine -gt
Snack) Conf(Wine -gt Snack)/P(Snack)
(2/3)/(5/6) 12/15 lt 1
17
2. Association Rules

• The Role of Lift
• Confidence just shows us a conditional
  probability
• Adjust Confidence by the frequency of the result
  item
• Lift Confidence/P(result)
• Avoid meaningless rules
• Supermarket vs Liquor Store?
• Promotion sales?
• Ex)
• Conf(A-gtB) 0.9
• If P(B) 1, then Lift(A-gtB) 0.9
• If P(B) 1/10, then Lift(A-gtB) 9

18
2. Association Rules

• Analyze the Result

19
2. Association Rules

• Analyze the Result
• Actionable Information
• Too trivial
• IF (Pizza) THEN (Coke)
• IF (Barbie Doll) THEN (Candy)
• So What
• Result of a previous action or a irregular event
• IF (Water) THEN (Chocolate)
• The dataset was gathered from 2/122/13

20
2. Association Rules

• Analyze the Result
• Correlation vs Causation
• Wine -gt Strong Heart
• Rice Noodle -gt Poor
• What can we do with the result?
• Changing display? (Shorter? Longer?)
• Discount?
• Cross-selling?
• Is there any profitable action?

21
2. Association Rules

• Analyze the Result
• Complement vs. Substitute
• Bread -gt Milk
• Beer -gt Wine
• Can we promote to cross-sell or discount?
• Why?

Vs.
22
2. Association Rules

• Conclusion
• Advantages
• Originality
• Interpretability
• Easy to explain to people (who issue grant)
• Actionable
• Difficulties
• Scalability (Min-Support concept with Apriori)
• Item-Transaction define
• Rare item
• Problem dependent

23

• Q A
• Homepage http//dmlab.snu.ac.kr
• E-mail dikim01_at_snu.ac.kr
• Phone [phone] (Lab.),
• 010-3439-7982 (Personal)