Data Mining

1
Data Mining

• Kelby Lee

2
Overview

• Transaction Database
• What is Data Mining
• Data Mining Primitives
• Data Mining Objectives
• Predictive Modeling
• Knowledge Discovery
• Other Objectives to Data Mining
• What Data Mining is Not
• Other Factors in Data Mining Categorization
• Conclusion

3
Transaction Database

• Relation Consisting of Transactions
• TID (Transaction Identifier)
• Regularities between Transaction Behavior

4
Transaction Database

• Table 1.1 Transaction Database
• TID Customer Item Date Price Quantity
• --------------------------------------------------
  --------------------------------------------------
  -----------------------------
• 100 C1 chocolate 01/11/2001 1.59 2
• 100 C1 ice cream 01/11/2001 1.89 1
• 200 C2 chocolate 01/12/2001 1.59 3
• 200 C2 candy bar 01/12/2001 1.19 2
• 200 C2 jackets 01/12/2001 120.39 2
• 300 C3 jackets 01/14/2001 168.88 1
• 300 C3 color shirts 01/14/2001 27.95 2
• 400 C4 jackets 01/15/2001 149.49 1

5
Association Rules

• A customer who buys chocolate will likely buy
  candy bar
• one type of Data Mining task

6
Discovered Rules

• Table 1.2 Discovered Rules
• Rule Bought this... ...also bought that
• --------------------------------------------------
  -----------------------------------------------
• 1 chocolate ice cream
• 2 candy bar chocolate
• 3 ski pants colored shirt
• 4 beer diaper

7
What is Data Mining

• Retrieve individual elements
• Given a name of a product, find price and
  producer
• Analysis
• Average monthly sales amount and derivation

8
Advances Allow For

• Large amounts of Data to be Handled
• Aspect of Analysis
• Data Rich but Knowledge Poor

9
Discover Patterns

• Improve Business Performance
• Exploit favorable patterns
• Avoid problematic patterns
• Increase Understanding
• Predict Outcome

10
Answer the Key Business Questions

• Who will buy? What will they buy? How much?
• Classification and Prediction
• What are the different types of Customers?
• Segmentation of Customers

11
Answer the Key Business Questions

• What relationship exists between customers or
  Website visitors and the products?
• Association
• What are the groupings hidden in the data?
• Clustering Analysis

12
Data Mining Definition

• Non Trivial Extraction of implicit, previously
  unknown, interesting, and potentially useful
  information from data

13
Different Types of Data Mining

• Business Data Mining
• Scientific Data Mining
• Internet Data Mining

14
Data Mining Applications

• Medical
• Control Theory
• Engineering
• Public Administration
• Marketing and Finance
• Data Mining on the Web
• Scientific Data Base
• Fraud Detection

15
Data Mining Primitives

• Fundamental Elements Needed to Define a Data
  Mining Task
• Eight Elements (P,D,K,B,T,M,I,U)
• 8 - Tuple

16
Elements

• P - Problem Specification
• D - Task Relevant Data
• K - Kind of Knowledge to be Mined
• B - Background Knowledge
• T - Specific algorithms or techniques
• M - Models developed or knowledge patterns
  extracted
• I - Interestingness
• U- User

17
Diagram
18
Relationship between Elements

• User Defines Problem (P) and specifies
  Interestingness (I)
• Data Miner with K and T as core elements
  utilizing D and B and incorporates I
• Data Miner produces M

19
Data Mining Objectives

• Discovery
• Finding human interpretable patterns describing
  the data
• Prediction
• Using some variables or fields in database to
  predict unknown or future values or other
  variables of interest

20
Data Mining Objectives

• Knowledge Discovery
• Stage somewhat prior to prediction where
  information is insufficient
• Closer to decision support

21
Predictive Modeling

• Predict Values Based on Similar Groups of Data
• Submit records with some unknown fields and
  system will predict value

22
Predictive Modeling

• Pattern Recognition
• Association of an observation to past experience
  or knowledge
• Interchangeable with classification

23
Predictive Modeling

• Classification
• Process of assigning finite set of labels to an
  observation
• Estimation
• Assign infinite number of numeric labels to an
  observation

24
Knowledge Discovery

• Find Patterns in Data Base
• If someone buys one thing, what else will they
  buy
• Interesting Certain Knowledge
• Output called Discovered Knowledge
• KDD - Knowledge Discovery in Data Base

25
Data Mining

• Is about why, about hidden regularities,
  important aspect related to perception, learning
  and evolving
• Decision support process in which we search
  patterns of information in data
• Once found, display in suitable format

26
Four Points of KDD

• Discovered Knowledge Represented in High-Level
  Language
• Accurately Portray contents of Database
• Interesting to user
• Process is Efficient

27
Important Issues

• Human Centered
• Under control of human user to meet human needs
• Incorporate Interestingness
• Provide Various Types
• Provide Visualization

28
Other Objectives

• Forensic analysis
• Applying extracted patterns to find anomalous or
  unusual data elements largely involved in
  business applications
• Find out what the norm is and find those that
  deviate from the norm

29
What Data Mining is Not

• Analysis vs Monitoring
• Analysis - previously collected information
• Monitoring
• Collect data as it comes in and compare to set of
  conditions
• Unexpected Discovery
• Must have general goal in mind

30
Other Factors in Categorization

• Data Retention
• Data is retained for future pattern matching
• Pattern Distillation
• Analyse data, extract pattern, leave data behind

31
Conclusion

• Transaction Database
• What is Data Mining
• Data Mining Primitives
• Data Mining Objectives
• Predictive Modeling
• Knowledge Discovery
• Other Objectives to Data Mining
• What Data Mining is Not
• Other Factors in Data Mining Categorization