Chapter 6 slides, Computer Networking, 3rd edition

1
Data Mining and Its Applications
Data Mining Techniques For Marketing, Sales,
and Customer Support, by Michael J.A. Berry and
Gordon Linoff, John Wiley Sons, Inc.,
1997. Discovering Data Mining from concept to
implementation, by Cabena, Harjinian, Stadler,
Verhees and Zanasi, Prentice Hall, 1997. Building
Data Mining Applications for CRM, by Alex Berson,
Stephen Smith and Kurt Thearling, McGraw Hall,
1999. Data Mining Cookbook Modeling Data for
Marketing, Risk, and Customer Relationship
Management, by Olivia Parr Rud, John Wiley
Sons, Inc, 2001. Mastering Data Mining The Art
and Science of Customer Relationship management,
by Michael J.A. Berry and Gordon S. Linoff, John
Wiley Sons, Inc, 2000. Machine Learning, by Tom
M. Mitchell, McGraw-Hill, 1997. Data Mining
Concepts and Techniques, by Jiawei Han and
Micheline Kamber, Morgan Kaufmann,
2001. Introduction to Data Mining, by Pang-Ning
Tan, Michael Steinbach, and Vipin Kumar, Addison
Wesley, 2005.
2
Why Mine Data?

• Lots of data is being collected and warehoused
• Web data, e-commerce
• purchases at department/grocery stores
• Bank/Credit Card transactions
• Computers have become cheaper and more powerful
• Competitive Pressure is Strong
• Provide better, customized services for an edge
  (e.g. in Customer Relationship Management)

3
Mining Large Data Sets - Motivation

• There is often information hidden in the data.
• Human analysts may take weeks to discover useful
  information.
• Much of the data is never analyzed at all.

The Data Gap
Total new disk (TB) since 1995
Number of analysts
4
What is Data Mining?

• Many Definitions
• Non-trivial extraction of implicit, previously
  unknown and potentially useful information from
  data
• Exploration analysis, by automatic or
  semi-automatic means, of large quantities of
  data in order to discover meaningful patterns

5
What is (not) Data Mining?

• What is Data Mining?
• Certain names are more prevalent in certain US
  locations (OBrien, ORurke, OReilly in Boston
  area)
• Group together similar documents returned by
  search engine according to their context (e.g.
  Amazon rainforest, Amazon.com,)

• What is not Data Mining?
• Look up phone number in phone directory
• Query a Web search engine for information about
  Amazon

6
Origins of Data Mining

• Draws ideas from machine learning/AI, pattern
  recognition, statistics, and database systems
• Traditional techniquesmay be unsuitable due to
• Enormity of data
• High dimensionality of data
• Heterogeneous, distributed nature of data

Statistics/AI
Machine Learning/ Pattern Recognition
Data Mining
Database systems
7
Data Mining Tasks

• Prediction Methods
• Use some variables to predict unknown or future
  values of other variables.
• Description Methods
• Find human-interpretable patterns that describe
  the data.

From Fayyad, et.al. Advances in Knowledge
Discovery and Data Mining, 1996
8
Data Mining Tasks...

• Classification
• Clustering
• Association Rule Discovery
• Sequential Pattern Discovery

9
The Virtuous Cycle of Data Mining
Identify business problems and
areas where analyzing data can
Act on the information
provide value
Measure the results of your efforts to provide
insight on how to
exploit your data.
Taken from a talk given by Michael J.A. Berry on
Data Mining for CRM.
10
Some Typical Business Problems

• Customer profiling
• Customer segmentation
• Direct marketing
• Customer retention
• Basket analysis (retail)
• Cross selling
• Fraud detection

11
Customer Profiling

• Question
• What kinds of customers were profitable in last
  year?
• Data
• Customer details such as Age, Gender, Occupation,
  Salary Levels, Account, etc.
• Earnings from customers in last year.
• Data Mining
• Divide customers into profitability categories
  according to earnings such as highly profitable,
  profitable, non-profitable, loss.
• Find rules using data mining techniques.
• Analyze the rules and take actions.

12
Customer Profiling Rules

• IF age gt 30 and Age lt45 and
• occupation is professional and
• salary level is between 50,000 and 70,000
• Then this user is profitable
• The rules are with some statistic support such as
  support and confidence.

13
Customer Segmentation

• Consumers are not same. They need to be treated
  differently. Segmentation is essential in
  marketing.
• Different spending capability
• Different spending potentials
• Different behaviors
• Different profitability
• Different preferences
• Different hobbies
• Different life style

14
Customer Segmentation

• Customer segmentation is a process to divide
  customers into different groups or segments.
  Customers in the same segment have similar needs
  or behaviors so that similar marketing strategies
  or service policies can be applied to them.
• Customer segments are required in several
  business areas including
• Marketing
• Customer services
• Products and service development
• Sales promotion
• Customer retention

15
Direct Marketing

• Question
• Select a customer mailing list for a product
  campaign.
• Purposes
• Reduce the campaign cost and obtain a high
  responding rate.
• Data
• Customer details and previous campaign data.

16
Life Cycle of a Loan Product
17
Business Objectives

• Mellon Bank Corporation is a major financial
  services company head-quarted in Pittsburgh.
• Build an extendible loan secured by the values of
  a clients own property.
• Achieve the highest possible Return On Investment
  (ROI).
• Based on customers with DDA, build a model for
  HELOC.

18
Data Preparaton

• The primary data source was the approximately
  40,000 Mellon customers who had (or once had)
  HELCOCs and DDAs.
• Data
• Demographic data sourced both internally and
  externally (age, income, length of residence, and
  other indicators of economic condition)
• DDA data (history of loan balance over 3, 6, 9,
  12, 18 months, history of returned checks,
  history of interest rates.
• Property data sourced externally (home purchase
  price, loan-to-value ratio)
• Other data related to credit worthiness
• Use 120 variables

19
(No Transcript)
20
Responders
21
Classification
22
Customer Retention

• Question
• Find out what kinds of customers tend to churn
  and build a model which can predict the
  likely-to-churn customers.
• Data mining solution
• Collect data about the customers who have
  churned.
• Select a set of customers who have been loyal.
• Merge the two data sets to form training, testing
  and evaluation data sets.

23
Basket Analysis
24
Basket Analysis
A
A
B
A
B
B
C
C
C
D
C
D
D
E
E
Rule A ? D C ? A A ? C B C ? D
Support 2/5 2/5 2/5 1/5
Confidence 2/3 2/4 2/3 1/3
25
Cross Selling Citicorp/Travelers Groups merger

• Online Newshour April 7, 1998 reports
• In the largest proposed corporate merger in
  history, the banking giant Citicorp and insurance
  titan Travelers will join forces. The new
  company, to be called Citigroup, would be the
  largest financial services company in the world
  
• One of the rationale for the merger
• MARCUS ALEXIS, Northwestern University commented
  
• Well, they have competitive issues. There
  are certain synergies. Not only do customers like
  to get a full range of services from a single
  vendor but also there are certain economies in
  cross selling by them.

(Online Newshour April 7, 1998)
26
Opportunities of Cross Selling

• Travelers Group can increase sales of insurance
  products from Citicorp customer base.
• Citicorp can increase sales of financial services
  from Travelers Group customer base.
• Customers get convenience by doing one stop
  shopping for both financial service and insurance
  products.

27
Cross Selling and up Selling

• Cross selling is the process of selling current
  customers new products after they purchased
  products of different categories.
• E.g., sell car maintenance products to customers
  who just bought new cars.
• Up selling is the process of selling current
  customers upgraded products or services after
  they purchased products of same category.
• E.g., Sell mobile voice service users data service

28
Understanding Customers
More Efficient Acquisition
Longer Lasting Relationship
More Frequent Up/Cross Sell
More Profit
Profit
Revenue
Less Loss
Time
Loss
Taken from SPSS talk.
29
Understanding Customers
More Efficient Acquisition
Longer Lasting Relationship
More Frequent Up/Cross Sell
Even More Profit
Profit
Revenue
Less Loss
Time
Loss
Taken from SPSS talk.
30
How Cross Selling Works

• Assume a marketing manager in a bank has the
  following products for customers
• Saving account
• Check account
• Standard credit card
• Gold credit card
• Primary mortgage
• Secondary mortgage
• The manager wants to design a new campaign to
  customers who
• Prepare to buy a new home
• Prepare to refinance an existing home
• Prepare to add a second mortgage

31
How to match customers with offers

• Determine three offers to customers
• New first mortgage
• Refinance of first mortgage
• Second mortgage
• Each customer is only made one offer

Customers
32
The Impact of Fraud

• GAO (The United States General Accounting Office)
  cited 19.1 billion in improper government
  payments in 17 major programs for fiscal year
  1998.
• Medicare 12.6 Billion
• Supplemental Security Income 1.6 B
• The Food Stamp Program 1.4 B
• Old Age and Survival Insurance 1.2 B
• Disability Insurance 941 Million
• Housing Subsidies 847 Million
• Veterans Benefits, Unemployment Insurance and
  Others 514 Million

33
Background

• HIC (The Health Insurance Commission) in
  Australia is a federal government agency.
• HIC pays insurance claims more than 20 million
  Australian dollars and pay out about A8 billion
  in funds every year.
• More than 300 million transactions are processed
  and stored every year. 1.3TB in five year.

34
Preventing Fraud and Abuse

• Business Objectives
• The focus of the HIC project was on the recent
  and steady 10 annual rise in the cost of
  pathology claims for clinical tests.
• Approaches
• To identify potential fraudulent claims or claims
  arising from inappropriate practice, and
• To develop general profiles of the GP practices
  in order to compare practice behaviors of
  individual GPs.

35
Data Proprocessing

• Two databases
• Episode Database
• One Episode record records a patient visit.
• In total, 6.8 million records.
• There were 227 different pathology tests.
• GP (doctor) database
• There are 17,000 records related to active GPs
• The behavior of 10,409 GPs was to be studied.
• A matrix of 10,409 by 227 elements.
• The elements were then scaled from 0 to 1 with
  respect to the total number of tests of each kind.

36
Input to Segmentation
37
Overview
38
Data Mining

• They conducted association rule mining, when
  support 0.25,the team decided that the
  presence of some tests in the input database was
  causing spurious rules to be revealed (Pathology
  Episode Initiation (PEI)).
• PEI tests depend on who ordered them and where
  they were ordered.
• When the PEI tests were removed, the number of
  rules dropped significantly.

39
Result Analysis

• A request for a microscopic examination of feces
  for parasites (OCP) was associated with a
  cultural examination of feces (FCS) in 0.85 of
  cases.
• A 92.6 chance that if OCP tests were requested,
  they would be done with FCS.
• A 0.61 of chance, OCP was associated with a
  different more expensive test called MCS32, which
  costs A13.55 per test.

40
GP Profiles
41
Discussions

• Segment 13
• Represent the majority of traditional GPs who
  are practicing conventionally. 5,450 GPs. Total
  52 of GPs.
• Only 6.2 of the medical pathology tests
• Segment 4
• 54 GPs. Only 0.51 of GPs.
• 2.7 of the medical pathology tests.

42
?? 2004.4.21
43
(No Transcript)