470643 ftoc.qxd 3/8/04 11:33 AM Page xi
Contents
xi
How Does a Neural Network Learn Using
Back Propagation?
228
Heuristics for Using Feed-Forward,
Back Propagation Networks
231
Choosing the Training Set
232
Coverage of Values for All Features
232
Number of Features
233
Size of Training Set
234
Number of Outputs
234
Preparing the Data
235
Features with Continuous Values
235
Features with Ordered, Discrete (Integer) Values
238
Features with Categorical Values
239
Other Types of Features
241
Interpreting the Results
241
Neural Networks for Time Series
244
How to Know What Is Going on Inside a Neural Network
247
Self-Organizing Maps
249
What Is a Self-Organizing Map?
249
Example: Finding Clusters
252
Lessons Learned
254
Chapter 8
Nearest Neighbor Approaches: Memory-Based
Reasoning and Collaborative Filtering
257
Memory Based Reasoning
258
Example: Using MBR to Estimate Rents in Tuxedo, New York
259
Challenges of MBR
262
Choosing a Balanced Set of Historical Records
262
Representing the Training Data
263
Determining the Distance Function, Combination
Function, and Number of Neighbors
265
Case Study: Classifying News Stories
265
What Are the Codes?
266
Applying MBR
267
Choosing the Training Set
267
Choosing the Distance Function
267
Choosing the Combination Function
267
Choosing the Number of Neighbors
270
The Results
270
Measuring Distance
271
What Is a Distance Function?
271
Building a Distance Function One Field at a Time
274
Distance Functions for Other Data Types
277
When a Distance Metric Already Exists
278
The Combination Function: Asking the Neighbors
for the Answer
279
The Basic Approach: Democracy
279
Weighted Voting
281
470643 ftoc.qxd 3/8/04 11:33 AM Page xii
xii
Contents
Collaborative Filtering: A Nearest Neighbor Approach to
Making Recommendations
282
Building Profiles
283
Comparing Profiles
284
Making Predictions
284
Lessons Learned
285
Chapter 9
Market Basket Analysis and Association Rules
287
Defining Market Basket Analysis
289
Three Levels of Market Basket Data
289
Order Characteristics
292
Item Popularity
293
Tracking Marketing Interventions
293
Clustering Products by Usage
294
Association Rules
296
Actionable Rules
296
Trivial Rules
297
Inexplicable Rules
297
How Good Is an Association Rule?
299
Building Association Rules
302
Choosing the Right Set of Items
303
Product Hierarchies Help to Generalize Items
305
Virtual Items Go beyond the Product Hierarchy
307
Data Quality
308
Anonymous versus Identified
308
Generating Rules from All This Data
308
Calculating Confidence
309
Calculating Lift
310
The Negative Rule
311
Overcoming Practical Limits
311
The Problem of Big Data
313
Extending the Ideas
315
Using Association Rules to Compare Stores
315
Dissociation Rules
317
Sequential Analysis Using Association Rules
318
Lessons Learned
319
Chapter 10 Link Analysis
321
Basic Graph Theory
322
Seven Bridges of Königsberg
325
Traveling Salesman Problem
327
Directed Graphs
330
Detecting Cycles in a Graph
330
A Familiar Application of Link Analysis
331
The Kleinberg Algorithm
332
The Details: Finding Hubs and Authorities
333
Creating the Root Set
333
Identifying the Candidates
334
Ranking Hubs and Authorities
334
Hubs and Authorities in Practice
336
470643 ftoc.qxd 3/8/04 11:33 AM Page xiii
Contents xiii
Case Study: Who Is Using Fax Machines from Home?
336
Why Finding Fax Machines Is Useful
336
The Data as a Graph
337
The Approach
338
Some Results
340
Case Study: Segmenting Cellular Telephone Customers
343
The Data
343
Analyses without Graph Theory
343
A Comparison of Two Customers
344
The Power of Link Analysis
345
Lessons Learned
346
Chapter 11 Automatic Cluster Detection
349
Searching for Islands of Simplicity
350
Star Light, Star Bright
351
Fitting the Troops
352
K-Means Clustering
354
Three Steps of the K-Means Algorithm
354
What K Means
356
Similarity and Distance
358
Similarity Measures and Variable Type
359
Formal Measures of Similarity
360
Geometric Distance between Two Points
360
Angle between Two Vectors
361
Manhattan Distance
363
Number of Features in Common
363
Data Preparation for Clustering
363
Scaling for Consistency
363
Use Weights to Encode Outside Information
365
Other Approaches to Cluster Detection
365
Gaussian Mixture Models
365
Agglomerative Clustering
368
An Agglomerative Clustering Algorithm
368
Distance between Clusters
368
Clusters and Trees
370
Clustering People by Age: An Example of
Agglomerative Clustering
370
Divisive Clustering
371
Self-Organizing Maps
372
Evaluating Clusters
372
Inside the Cluster
373
Outside the Cluster
373
Case Study: Clustering Towns
374
Creating Town Signatures
374
The Data
375
Creating Clusters
377
Determining the Right Number of Clusters
377
Using Thematic Clusters to Adjust Zone Boundaries
380
Lessons Learned
381
470643 ftoc.qxd 3/8/04 11:33 AM Page xiv
xiv
Contents
Chapter 12 Knowing When to Worry: Hazard Functions and
Survival Analysis in Marketing
383
Customer Retention
385
Calculating Retention
385
What a Retention Curve Reveals
386
Finding the Average Tenure from a Retention Curve
387
Looking at Retention as Decay
389
Hazards 394
The Basic Idea
394
Examples of Hazard Functions
397
Constant Hazard
397
Bathtub Hazard
397
A Real-World Example
398
Censoring 399
Other Types of Censoring
402
From Hazards to Survival
404
Retention 404
Survival 405
Proportional Hazards
408
Examples of Proportional Hazards
409
Stratification: Measuring Initial Effects on Survival
410
Cox Proportional Hazards
410
Limitations of Proportional Hazards
411
Survival Analysis in Practice
412
Handling Different Types of Attrition
412
When Will a Customer Come Back?
413
Forecasting 415
Hazards Changing over Time
416
Lessons Learned
418
Chapter 13 Genetic Algorithms
421
How They Work
423
Genetics on Computers
424
Selection 429
Crossover 430
Mutation 431
Representing Data
432
Case Study: Using Genetic Algorithms for
Resource Optimization
433
Schemata: Why Genetic Algorithms Work
435
More Applications of Genetic Algorithms
438
Application to Neural Networks
439
Case Study: Evolving a Solution for Response Modeling
440
Business Context
440
Data 441
The Data Mining Task: Evolving a Solution
442
Beyond the Simple Algorithm
444
Lessons Learned
446
470643 ftoc.qxd 3/8/04 11:33 AM Page xv
Contents
xv
Chapter 14 Data Mining throughout the Customer Life Cycle
447
Levels of the Customer Relationship
448
Deep Intimacy
449
Mass Intimacy
451
In-between Relationships
453
Indirect Relationships
453
Customer Life Cycle
454
The Customer’s Life Cycle: Life Stages
455
Customer Life Cycle
456
Subscription Relationships versus Event-Based Relationships
458
Event-Based Relationships
458
Subscription-Based Relationships
459
Business Processes Are Organized around
the Customer Life Cycle
461
Customer Acquisition
461
Who Are the Prospects?
462
When Is a Customer Acquired?
462
What Is the Role of Data Mining?
464
Customer Activation
464
Relationship Management
466
Retention 467
Winback 470
Lessons Learned
470
Chapter 15 Data Warehousing, OLAP, and Data Mining
473
The Architecture of Data
475
Transaction Data, the Base Level
476
Operational Summary Data
477
Decision-Support Summary Data
477
Database Schema
478
Metadata 483
Business Rules
484
A General Architecture for Data Warehousing
484
Source Systems
486
Extraction, Transformation, and Load
487
Central Repository
488
Metadata Repository
491
Data Marts
491
Operational Feedback
492
End Users and Desktop Tools
492
Analysts 492
Application Developers
493
Business Users
494
Where Does OLAP Fit In?
494
What’s in a Cube?
497
Three Varieties of Cubes
498
Facts 501
Dimensions and Their Hierarchies
502
Conformed Dimensions
504
470643 ftoc.qxd 3/8/04 11:33 AM Page xvi
xvi
Contents
Star Schema
505
OLAP and Data Mining
507
Where Data Mining Fits in with Data Warehousing
508
Lots of Data
509
Consistent, Clean Data
510
Hypothesis Testing and Measurement
510
Scalable Hardware and RDBMS Support
511
Lessons Learned
511
Chapter 16 Building the Data Mining Environment
513
A Customer-Centric Organization
514
An Ideal Data Mining Environment
515
The Power to Determine What Data Is Available
515
The Skills to Turn Data into Actionable Information
516
All the Necessary Tools
516
Back to Reality
516
Building a Customer-Centric Organization
516
Creating a Single Customer View
517
Defining Customer-Centric Metrics
519
Collecting the Right Data
520
From Customer Interactions to Learning Opportunities
520
Mining Customer Data
521
The Data Mining Group
521
Outsourcing Data Mining
522
Outsourcing Occasional Modeling
522
Outsourcing Ongoing Data Mining
523
Insourcing Data Mining
524
Building an Interdisciplinary Data Mining Group
524
Building a Data Mining Group in IT
524
Building a Data Mining Group in the Business Units
525
What to Look for in Data Mining Staff
525
Data Mining Infrastructure
526
The Mining Platform
527
The Scoring Platform
527
One Example of a Production Data Mining Architecture
528
Architectural Overview
528
Customer Interaction Module
529
Analysis Module
530
Data Mining Software
532
Range of Techniques
532
Scalability 533
Support for Scoring
534
Multiple Levels of User Interfaces
535
Comprehensible Output
536
Ability to Handle Diverse Data Types
536
Documentation and Ease of Use
536
470643 ftoc.qxd 3/8/04 11:33 AM Page xvii
Contents xvii
Availability of Training for Both Novice and
Advanced Users, Consulting, and Support
537
Vendor Credibility
537
Lessons Learned
537
Chapter 17 Preparing Data for Mining
539
What Data Should Look Like
540
The Customer Signature
540
The Columns
542
Columns with One Value
544
Columns with Almost Only One Value
544
Columns with Unique Values
546
Columns Correlated with Target
547
Model Roles in Modeling
547
Variable Measures
549
Numbers 550
Dates and Times
552
Fixed-Length Character Strings
552
IDs and Keys
554