<<

. 130
( 137 .)



>>


marriages

294“295

categorical values, 239“240

purchases, 289

house-hold level data, 96

support, 301

mass intimacy, customer relationships,

telecommunications customers, 288

451“453

time attributes, 293

massively parallel processor

market research

(MPP), 485

control group response versus, 38

maximum values, of simple functions,

literature, 22

generic algorithms, 424

shortcomings, 25

MBR. See memory-based reasoning

survey-based, 113

MDL (minimum description

marketing campaigns. See also
length), 78

advertising

mean between time failure

acquisitions-time data, 108“110

(MTBF), 384

canonical measurements, 31

mean time to failure (MTTF), 384

champion-challenger approach, 139

mean values, statistics, 137

credit risks, reducing exposure to,

measurement errors, 159

113“114

median customer lifetime value,

cross-selling, 115“116

retention, 387

customer response, tracking, 109

median values, statistics, 137

customer segmentation, 111“113

medical insurance claims, useful

differential response analysis,

data sources, 60

107“108

medical treatment applications,

discussed, 95

MBR, 258

fixed budgets, 97“100

meetings, brainstorming, 37

loyalty programs, 111

memory-based reasoning (MBR)

new customer information,

case study, 259“262

gathering, 109“110

challenges of, 262“265

people most influenced by, 106“107

classification codes, 266, 273“274

planning, 27

combination function, 258, 265

profitability, 100“104

customer classification, 90“91

proof-of-concept projects, 600

customer response prediction, 258

response modeling, 96“97

632 Index


memory-based reasoning (MBR) missing data

(continued) data correction, 73“74

democracy approach, 279“281
NULL values, 590

distance function, 258, 265, 271“272
splits, decision trees, 174“175

fraud detection, 258
mission-critical applications, 32

free text response, 258
mode values, statistics, 137

historical records, selecting, 262“263
models

medical treatment applications, 258
assessing

new customers, 277
classifiers and predictors, 79

relevance feedback, 267“268
descriptive models, 78

similarity measurements, 271“272
directed models, 78“79

training data, 263“264
estimators, 79“81

weighted voting, 281“282
building, 8, 77





Y
men, differential response analysis comparing, using lift ratio, 81“82

and, 107
deploying, 84“85





FL
messages, prospecting, 89“90
model sets

metadata repository, 484, 491
balanced datasets, 68

AM
methodologies
components of, 52

data correction, 72“74
customer signatures, assembling, 68

data exploration, 64“68
partitioning, 71“72

data mining process, 54“55
predictive models, 70“71

TE

data selection, 60“64
timelines, multiple, 70

data transformation, 74“76
non-response, mass mailings, 35

data translation, 56“60
score sets, 52

learning sources
motor vehicle registration records,

truthful, 48“50 useful data sources, 61

untruthful, 44“48
MOU (minutes of use), wireless

model assessment, 78“82
communications industries, 38

model building, 77
MPP (massively parallel processor), 485

model deployment, 84“85
MSA (metropolitan statistical area), 94

model sets, creating, 68“72
MTBF (mean between time failure), 384

reasons for, 44
MTTF (mean time to failure), 384

results, assessing, 85
multiway splits, decision trees, 171

metropolitan statistical area (MSA), 94
mutation, generic algorithms, 431“432

minimum description length

N
(MDL), 78

N variables, dimension, 352

minimum support pruning, decision

National Consumer Assets Group

trees, 312

(NCAG), 23

minutes of use (MOU), wireless

natural association, automatic cluster

communications industries, 38

detection, 358

misclassification rates, binary

classification, 98


<<

. 130
( 137 .)



>>