A
accuracy criterion 193–194
acquisition cost 417–418
activation functions
about 243, 322
output layer 247
target layer 270–272
Add value 306
adjusted frequencies 441
adjusted probabilities, expected profits using 236
AIC (Akaike Information Criterion) 350–352
Append node 48–50, 116
Arc Tanget function 243–244
Architecture property
about 316
MLP setting 247
Neural Network node 281, 283, 284, 293, 295–297, 298–300, 305
NRBFUN network 303–304
Regression node 389
architectures
alternative built-in 286–307
of neural networks 316
user-specified 305–307
Assessment Measure property 174, 187, 193, 198, 387–389, 396–399
attrition, predicting 384–392
auto insurance industry, predicting risk in 3–4
AutoNeural node 307–309, 314–315, 316
Average method 408
average profit, vs. total profit for comparing tree size 192–193
average squared error 174, 194
B
β, as vector of coefficients 322
Backward Elimination method
about 335
when target is binary 335–337
when target is continuous 338–340
bank deposit products, predicting rate sensitivity of 4–5
bin
See groups
binary split search, splitting nodes using 176–177
binary targets
Backward Elimination method with 335–337
Forward Selection method with 340–342
models for 384–392
with nominal-scaled categorical inputs 135–138
with numeric interval-scaled inputs 129–135
regression models with 321–324
stepwise selection method with 344–345
binning transformations 96–97
Bonferroni Adjustment property 183–184
Boolean retrieval method 427
branch 170
bucket 96
See also groups
business applications
logistic regression for predicting mail campaign response 359–371
of regression models 358–379
C
calculating
Chi-Square statistic for continuous input 113–115
cluster components 64
Cramer's V for continuous input 113–115
eigenvectors 111
misclassification rate/accuracy rate 193–194
principal components 112
residuals 403–404
validation profits 190–192
worth of a tree 173–175
worth of splits 177–182
categorical variables 1–2, 165
child nodes 170
Chi-Square
calculating for continuous input 113–115
criterion for 130–135, 137–138
selection method 73
statistic 52, 53
test for 234
Chi-Square property, StatExplore node 114
class inputs, transformations of 98
Class Inputs property, Transform Variables node 98, 99, 162, 376
class interval
See groups
Class Levels Count Threshold property 19, 20, 108, 109, 164, 250, 267
Cloglog 335
Cluster Algorithm property 451–458
Cluster node 50, 69–72, 451
Cluster Variable Role property, Cluster node 70, 72
Clustering Source property, Variable Clustering node 63
clusters and clustering
assigning variables to 64–65
EM (Expectation-Maximization) 452–458, 460–461
hierarchical 451–459
selecting components 148–150
selecting variables for 140–148
Code Editor property, SAS Code node 103–104
combination functions 243, 270–272
combining
groups 88–90
models 383–413
predictive models 402–411
comparing
alternative built-in architectures of neural networks 286–307
categorical variables with ungrouped variables 165
gradient boosting and ensemble methods 410–411
models 383–413
models generated by DMNeural, AutoNeural, and Dmine Regression nodes 314–315
samples and targets 8
Complementary Log-Log link (Cloglog) 335
continuous input, calculating Chi-Square and Cramer's V for 113–115
continuous targets
Backward Elimination method with 338–340
with Forward Selection method 342–343
with nominal-categorical inputs 124–129
with numeric interval-scaled inputs 119–124
regression for 371–379
regression models with 333
stepwise selection method with 345–347
Correlations property, StatExplore node 55
Cosine function 243
cost of default 418–419
Cramer's V 53–54, 113–115
Cross Validation Error 355
Cross Validation Misclassification rate 355
Cross Validation Profit/Loss criterion 357–358
customer attrition, predicting 6
customer lifetime value 422
customer profitability
about 415–417
acquisition cost 417–418
alternative scenarios of response and risk 422
cost of default 418–419
customer lifetime value 422
extending results 423
optimum cut-off point 421–422
profit 419–421
revenue 419
Cutoff Cumulative property, Principal Components node 92–93
Cutoff Value property
Replacement node 81
Transform Variables node 98
cut-off values 258
D
data
applying decision tree models to prospect 173
pre-processing 8–10
data cleaning 9
data matrix 427–428, 430–431
Data Mining the Web (Markov and Larose) 458
data modification, nodes for
Drop 10, 79–80
Impute 10, 83, 153–154, 360, 386
Interactive Binning 83–90
Principal Components 90–95
Replacement 80–83
Transform Variables (See Transform Variables node)
Data Options dialog box 58
Data Partition node
about 27–28, 29, 249, 250
loss frequency as an ordinal target 269
Partitioning Method property 28, 386
property settings 197
Regression node 360, 372
variable selection 139, 145
variable transformation 153–154
Data Set Allocations property, Data Partition node 28
data sets
applying decision tree models to score 205–208
creating from text files 433–435
scoring using Neural Network models 263–266
scoring with models 277–279
Data Source property, Input Data node 26–27
data sources
changing measurement scale of variables in 164–165
creating 16–25, 37–40, 436
creating for text mining 436
creating for transaction data 37–40
decision 171, 174
decision tree models
about 170–172
accuracy/misclassification criterion 193–194
adjusting predicted possibilities for over-sampling 235–236
applying to prospect data 173
assessing using Average Square Error 194
average profit vs. total profit 192–193
binary split searches 176–177
calculating worth of trees 173–175
compared with logistic regression models 172
controlling growth of trees 185
developing interactively 215–233
developing regression tree model to predict risk 208–215
exercises 236–237
impurity reduction 182–183
measuring worth of splits 177–182
Pearson's Chi-square test 234
for predicting attrition 387–389
predicting response to direct marketing with 195–208
for predicting risk in auto insurance 396–399
pruning trees 185–192
p-value adjustment options 183–185
regression tree 176
roles of training and validation data in 175
in SAS Enterprise Miner 176–195
selecting size of trees 194–195
Decision Tree node
See also decision trees
about 117–118, 163
bins in 97
building decision tree models 187, 195
Interactive property 222, 225, 231–233
Leaf Role property 151
logistic regression 366, 367, 368, 377
in process flow 134
regression models 387–389, 392–401
Regression node 359
variable selection in 121
variable selection using 150–153
decision trees
developing interactively 215–233
growing 233
Decision Weights tab 22–23
Decisions property 187
Decisions tab 23
Default Filtering Method property, Filter node 29
Default Input Method property, Impute node 83
Default Limits Method property, Replacement node 81
default methods 98–100
degree of separation 178–179
depth adjustment 184
depth multiplier 184
Diagram Workspace 15, 16
dimension reduction 429–431
direct mail, predicting response to 2–3
direct marketing, predicting response to 195–208
DMDB procedure 78–79
Dmine Regression node 312–313, 314–315, 316
DMNeural node 309–312, 314–315, 316
documents, retrieving from World Wide Web 432–433
document-term matrix 427–428
Drop from Tables property, Drop node 80
Drop node 10, 79–80
E
EHRadial value 307
Eigenvalue Source property 91
eigenvalues 64, 110–115
eigenvectors 110–115
Elliot function 243–244
EM (Expectation-Maximization) clustering 452–458, 460–461
Ensemble node 384, 402, 407–409, 410–411
Entropy 180–181
Entry Significance Level property, Regression node
Forward Selection method 340–342, 342–343
regression models 372, 386
Stepwise Selection method 345–347
EQRadial value 307
EQSlopes value 307
error function 248
EVRadial value 307
EWRadial value 307
exercises
decision tree models 236–237
models, combining 412–413
neural network models 318–319
predictive modeling 115–116
regression models 382
textual data, predictive modeling with 461
variable selection 166–167
Expectation-Maximization (EM) clustering 452–458, 460–461
expected losses 423
expected lossfrq 394
explanatory variables 170, 241
Exported Data property
Input Data node 102
Time Series node 43–44
F
false positive fraction 258
File Import node 32–35
Filter node 10, 28–32, 445
Filter Viewer property 445
fine tuning 27
Forward Selection method
about 340
when target is binary 340–342
when target is continuous 342–343
frequency
about 8
adjusted 441
FREQUENCY procedure 54
frequency weighting 440, 441
Frequency Weighting property 444
G
Gauss function 243
Gini Cutoff property, Interactive Binning node 84–85
Gini Impurity Index 180
gradient boosting 402–404
Gradient Boosting node 402, 404–406, 410–411
GraphExplore node 50, 51, 58–61
groups
See also leaf nodes
combining 88–90
splitting 85–88
H
Help Panel 15
Hidden Layer Activation Function property 241, 281, 283, 287–288, 305
Hidden Layer Combination Function property 241, 281, 283, 287–288, 305–306
hidden layers 242–246
Hide property
Regression node 363
Transform Variables node 101, 156, 159, 162
transforming variables 162
Hide Rejected Variables property, Variable Selection node 120, 122
hierarchical clustering 451–459
Huber-M Loss 411
Hyperbolic Tangent function 243–244
I
Identity link 335
Import File property, File Import node 33, 34
Imported Data property, SAS Code node 102–103
impurity reduction
about 53
as measure of goodness of splits 179–180
when target is continuous 182–183
Impute node 10, 83, 153–154, 360, 386
Include Class Variables property, Variable Clustering node 139–140
initial data exploration, nodes for
about 50–51
Cluster 50, 69–72, 451
Graph Explore 50, 51, 58–61
MultiPlot 50, 51, 56–58, 358
Stat Explore 10, 50, 51–56, 79, 114, 358
Variable Clustering 50, 61–69, 117–118, 138–150, 163, 352
Variable Selection 50, 72–79, 153–154, 155–157, 163, 164, 359
input 111, 170
Input Data node
about 10, 249, 250
building decision tree models 195–196, 205
Data Source property 26–27
Exported Data property 102
loss frequency as an ordinal target 267
in process flow 91, 101
regression models 385–386, 407
scoring datasets 277
transforming variables 153–154
Input Data Source node 333, 360, 372
input layer 242
Input Standardization property 305
input variables, regression with large number of 11
inputs window 6
Interactive Binning node 83–90
Interactive Binning property, Interactive Binning node 85, 88–89
Interactive property 216
Interactive property, Decision Tree node 222, 225, 231–233
Interactive Selection property, Principal Components node 91, 93
intermediate nodes 130
Interval Criterion property 177–178, 182, 183
interval inputs, transformations for 95–98
Interval Inputs property
Merge node 45–46, 47
Regression node 364
Transform Variables node 95, 96, 98, 155, 159, 162
interval variables 2
Interval Variables property
Filter node 30
StatExplore node 52, 55, 114
inverse link function 322
K
KeepHierarchies property, Variable Clustering node 65
L
Larose, D.T.
Data Mining the Web 458
latent semantic indexing 429–431
leaf nodes 170, 233
See also terminal nodes
Leaf Role property
Decision Tree node 151
Regression node 377
Leaf Rule property 389–390
Leaf Size property 185, 215, 367
Leaf Variable property 389–390
Least Absolute Deviation Loss 411
Least Squares Loss 411
lift 174–175
lift charts 393–394
Linear Combination function 305–306
Linear Regression 333
Linear value 306
link function 321
Link Function property, Regression node 324, 333–335, 348
Logistic function 244
logistic regression
about 333
for predicting attrition 386–387
for predicting mail campaign response 359–371
with proportional odds 394–396
logistic regression models, vs. decision tree models 172
Logit link 322, 334
Logit Loss 412
logworth 178–179
loss frequency 208, 240, 266–279
M
marginal profit 420
marginal revenue 420
Markov, Z.
Data Mining the Web 458
maximal tree 175, 191–192
Maximum Clusters property, Variable Clustering node 62
Maximum Depth property 185, 215, 367
Maximum Eigenvalue property, Variable Clustering node 62, 64
Maximum method 408
Maximum Number of Steps property 345–347
maximum posterior probability/accuracy, classifying nodes by 193
measurement scale 1–2, 107–109
measurement scale, of variables 164–165
Menu Bar 15
Merge node 45–47, 48, 159–162
Merging property, Transform Variables node 161
Metadata Advisor Options window 19
Method property 84, 187, 367
methods
Average 408
Backward Elimination 335–340
Boolean retrieval 427
Chi-Square selection 73
default 98–100
frequency weighting 441
Maximum 408
R-Square selection 72–73
term weighting 441–445
Minimum Chi-Square property, Variable Selection node 73
Minimum property, Cluster node 70
Minimum R-Square property, Variable Selection node 73, 74, 120–121, 157
misclassification criterion 174, 193–194
MLP (Multilayer Perception) neural network 279–281, 287–288
Model Comparison node
assessing predictive performance of estimated models 254–258
building decision tree models 204–205, 211, 233
comparing alternative built-in architectures 286
in process flow 139, 149
regression models 391
Regression node 360, 368
variable selection 151
Model Selection Criterion property 248, 249, 250, 273, 389
Model Selection property 386
modeling data, sources of 8
modeling strategies, alternative 10–11
models
See also neural network models
for binary targets 384–392
combining 412–413
comparing and combining 383–413
for ordinal targets 392–401
Multilayer Perception (MLP) neural network 279–281, 287–288
Multiple Method property, Transform Variable node 162
MultiPlot node 50, 51, 56–58, 358
N
Network property, Neural Network node 251, 269, 283
neural network models
about 240–241
alternative specifications of 279–286
AutoNeural node 314–315
comparing alternative built-in architectures of Neural Network node 286–309
Dmine Regression node 312–313, 314–315
DMNeural node 309–312, 314–315
estimating weights in 247–249
exercises 318–319
general example of 241–247
nodes for 240–241
for predicting attrition 389–392
predicting loss frequency in auto insurance 266–279
for predicting risk in auto insurance 400–401
scoring data sets using 263–266
target variables for 240
Neural Network node
about 240–241, 316
Architecture property 281, 283, 284, 293, 295–297, 298–300, 305
loss frequency as an ordinal target 268–269
Model Selection Criterion property 273
Multilayer Perceptron (MLP) neural networks 278–281
Normalized Radial Basis Function with Equal Heights and Unequal Widths (NRBFEH) 295–297
Normalized Radial Basis Function with Equal Volumes (NRBFEV) 301–302
Normalized Radial Basis Function with Equal Widths and Heights (NRBFEQ) 292–294
Ordinary Radial Basis Function with Equal Heights and Unequal Widths (ORBFUN) 291–292
Radial Basis Function neural networks in 282–286
regression models 389–390, 392–401
score ranks in Results window 275
scoring datasets 277
selecting optimal weights 261–263
setting properties of 250–254
target layer combination and activation functions 270–272
neural networks
about 316
alternative specifications of 279–286
comparing alternative built-in architectures in 286–307
node definitions 175
Node (Tool) group tabs 15
Node ID property, Transform Variables node 159, 161
nodes
See also Data Partition node
See also Decision Tree node
See also Input Data node
See also Model Comparison node
See also Neural Network node
See also Regression node
See also SAS Code node
See also Transform Variables node
See also Variable Clustering node
See also Variable Selection node
Append 48–50, 116
AutoNeural node 307–309, 314–315, 316
child 170
classifying by maximum posterior probability/accuracy 193
Cluster 50, 69–72, 451
for data modification 79–101
Dmine Regression 312–313, 314–315, 316
DMNeural 309–312, 314–315, 316
Drop 10, 79–80
Ensemble 384, 402, 407–409, 410–411
File Import 32–35
Filter 10, 28–32, 445
Gradient Boosting 402, 404–406, 410–411
GraphExplore 50, 51, 58–61
Impute 10, 83, 153–154, 360, 386
for initial data exploration 50–79
Input Data Source 333, 360, 372
Interactive Binning 83–90
intermediate 130
leaf 170, 233 (See also terminal nodes)
Merge 45–47, 48, 159–162
MultiPlot 50, 51, 56–58, 358
for neural network models 240–241
parent 170
Principal Components 90–95
Replacement 80–83
responder 192
Root 130, 170, 225–231
sample 26–50
Score 205–207, 265, 266, 277
splitting using binary split search 176–177
StatExplore 10, 50, 51–56, 79, 114, 358
Stochastic Boosting 384
terminal 130, 170
Text Cluster 450, 451–458, 458–461
Text Filtering 432, 440–450, 442
Text Import 431, 436
Text Parsing 431, 436–440, 442, 445–450
Text Topic 445–450
Time Series 35–44
Transformation 376
utility 101–107
nominal categorical (unordered polychotomous) target, predicting 7
Nominal Criterion property 177, 180–181
nominal (unordered) target, regression models with 329–332
nominal-categorical inputs, continuous target with 124–129
nominal-scaled categorical inputs, binary target with 135–138
non-responders 234
NRBFEH (Normalized Radial Basis Function with Equal Heights and Unequal Widths) 295–297
NRBFEQ (Normalized Radial Basis Function with Equal Widths and Heights) 292–294
NRBFEV (Normalized Radial Basis Function with Equal Volumes) 300–302
NRBFEW (Normalized Radial Basis Function with Equal Widths and Unequal Heights) 297–300
NRBFUN (Normalized Radial Basis Function with Unequal Widths and Heights) 302–304
Number of Bins property
about 131
StatExplore node 52
Variable Selection node 73
Number of Hidden Units property 279–281, 286, 389, 400–401
number of levels, of variables 107–109
numeric interval-scaled inputs
binary target with 129–135
continuous target with 119–124
O
observation weights 8
observed proportions 170
Offset Value property, Transform Variables node 96
opening SAS Enterprise Miner 12.1 14
operational lag 6
optimal binning 45, 96–97
optimal tree 175
Optimization property 250
optimum cut-off point 421–422
ORBFEQ (Ordinary Radial Basis Function with Equal Heights and Widths) 288–290
ORBFUN (Ordinary Radial Basis Function with Equal heights and Unequal Widths) 291–292
ORBFUN (Ordinary Radial with Unequal Widths) 283
ordered polychotomous targets
See ordinal targets
Ordinal Criterion property 177, 180–181
ordinal targets
loss frequency as 267–279
models for 392–401
regression models with 324–329
original segment 170
output data sets
created by Time Series node 43–44
developing predictive equations created by Text Topic node 449–450
output layer 246–247
overriding default methods 99–100
over-sampling, adjusting predicted probabilities for 235–236
P
p weights 284
parent nodes 170
Partitioning Method property, Data Partition node 28, 386
Pearson Correlations property, StatExplore node 55
Pearson's Chi-square test 234
percentage of ranked data (n%) 175
performance window 6, 7
posterior probability
about 170
for leaf nodes from training data 189
of non-response 203
of response 203
Posterior Probability property 408
Predicted Values property 408
predicting
attrition 384–392
customer attrition 6
loss frequency in auto insurance with Neural Network model 266–279
nominal categorical (unordered polychotomous) target 7
rate sensitivity of bank deposit products 4–5
response (See neural network models)
response to direct mail 2–3
response to direct marketing 195–208
risk (See neural network models)
risk in auto insurance industry 3–4
risk of accident risk 392–401
risk with regression tree models 208–215
predictive equations, developing using output data set created by Text Topic node 449–450
predictive modeling
See also textual data, predictive modeling with
about 14
boosting 402–411
combining 402–411
creating new projects in SAS Enterprise Miner 12.1 14–15
creating process flow diagrams 25–26
creating SAS data sources 16–25
eigenvalues 64, 110–115
eigenvectors 110–115
exercises 115–116
measurement scale 107–109
nodes for data modification 79–101
nodes for initial data exploration 50–79
number of levels of variable 107–109
opening SAS Enterprise Miner 12.1 14
principal components 110–115
sample nodes 26–50
SAS Enterprise Miner window 15–16
type of variable 107–109
utility nodes 101–107
Preliminary Maximum property, Cluster node 70
pre-processing data 8–10
principal components 110–115
Principal Components node 90–95
Prior Probabilities tab 22
probabilities, adjusted 236
Probit link 334
process flow diagrams 25–26, 40
profit 419–421
See also customer profitability
See also validation profit
average vs. total 192–193
marginal 420
Profit/Loss criterion 357
Project Panel 15
projects, creating in SAS Enterprise Miner 12.1 14–15
promotion window 4
properties
See also specific properties
of Neural Network node 250–254
of Regression node 333–358
Properties Panel 15
Proportional Odds model 394–396
pruning trees 175, 185
p-value 52
P-value adjustment options
Bonferroni Adjustment property 183–184
depth adjustment 184
Leaf Size property 185
Split Adjustment property 184
Threshold Significance Level property 184
Time of Kass Adjustment property 184
Q
quantifying textual data 426–428, 431–432
quantile 96
R
rate sensitivity, predicting of bank deposit products 4–5
RBF (Radial Basis Function) neural network 281–286
Receiver Operating Characteristic (ROC) charts 258–261
recursive partitioning 130, 170
regression
for continuous targets 371–379
with large number of input variables 11
regression models
about 321
with binary targets 321–324
business applications 358–379
exercises 382
Regression node properties 333–358
types of models developed using 321–333
Regression node
See also regression models
about 10, 11
Architecture property 389
Chi-Square criterion 138
Data Partition node 360, 372
Decision Tree node 359
Entry Significance Level property 340–342, 342–343, 345–347, 372, 386
Hide property 363
Interval Inputs property 364
Leaf Role property 377
Link Function property 324, 333–335, 348
predictive modeling 449–450
in process flow 76, 90, 95, 96, 101, 123–124, 129, 133, 139, 143–144
properties of 333–358
regression models 386, 387–389, 392–401
Regression Type property 324, 333, 348
Reject property 363
R-Square criterion 130, 136–137
Selection Model property 144, 335–347, 340, 342–343, 348, 367, 394, 449–450
testing significance of dummy variables 98
testing variables and transformations 45, 47, 48
Transform Variables node 359
transforming variables 157, 159, 161–162
Variable Clustering node 352
variable selection 145, 146, 148–149, 150, 151, 152
variable selection in 121
Variable Selection property 377, 389–390
Variables property 95, 123–124, 133–134
regression tree 176, 208–215
Regression Type property, Regression node 324, 333, 348
Reject property
Regression node 363
Transform Variables node 101, 156, 159, 162
transforming variables 162
Replacement Editor property, Replacement node 81–82
Replacement node 80–83
research strategy
about 1
alternative modeling strategies 10–11
defining targets 2–8
measurement scales for variables 1–2
pre-processing data 8–10
residuals, calculating 403–404
responder node 192
responders 234
response
alternative scenarios of 422
predicting (See neural network models)
predicting to direct mail 2–3
revenue 419
risk
See also neural network models
alternative scenarios of 422
classifying for rate setting 279
predicting in auto insurance industry 3–4
predicting with regression tree models 208–215
risk rate 415–416
ROC (Receiver Operating Characteristic) charts 258–261
Role property 263
Root node 130, 170, 225–231
R-Square criterion 130, 136–137
R-Square selection method 72–73
S
sample nodes
Append 48–50, 116
Data Partition 27–28, 29, 139, 145, 153–154, 198, 249, 250, 269, 360, 372
File Import 32–35
Filter 10, 28–32, 445
Input Data 10, 26–27, 91, 101, 102, 153–154, 195–196, 205, 249, 250, 267, 277, 385–386, 407
Merge 45–47, 48, 159–162
Time Series 35–44
samples, compared with targets 8
SAS Code node
about 10, 101–107
building decision tree models 207–208
logistic regression 374
predictive modeling 438, 444
score ranks in Results window 275
SAS Enterprise Miner
creating projects in 14–15
data cleaning after launching 9
data cleaning before launching 9
developing decision trees in 176–195
opening 14
window 15–16
SAS Enterprise Miner: Reference Help 284, 437
SBC (Schwarz Bayesian Criterion) 352–353
Score node 205–207, 265, 266, 277
scoring
data sets using Neural Network models 263–266
datasets with models 277–279
showing ranks in Results window 273–276
segments 170
See also leaf nodes
Select an Analysis property, Time Series node 41
Selection Criterion property, Regression node
about 348–350
Akaike Information Criteria (AIC) 350–352
Backward Elimination method 335
cross validation error 355
cross validation misclassification rate 355
Cross Validation Profit/Loss Criterion 357–358
Forward Selection method 341–342, 342–343
logistic regression 394
predictive modeling with textual data 449–450
Profit/Loss Criterion 357
regression models 365, 367, 376, 386
Schwarz Bayesian Criterion (SBC) 352–353
validation error 353–354
validation misclassification 354
Validation Profit/Loss Criterion 355–356
variable selection 144
Selection Default property 343
Selection Model property, Regression node 144, 335–347, 340, 342–343, 348, 367, 394, 449–450
Selection Options property 336, 344–345
sensitivity
See true positive fraction
separation, degree of 178–179
Significance Level property 184, 215, 348, 367
simple transformation 96
Sine function 243
Singular Value Decomposition (SVD) 429, 431
sources, of modeling data 8
Spearman Correlations property, StatExplore node 55
specificity
See true positive fraction
Split Adjustment property 184, 215
split point, changing of nominal variables 222–225
Split Size property 185
splits, measuring worth of 177–181, 181–182
splitting
groups 85–88
nodes using binary split search 176–177
process of 62
Splitting Rule Criterion property 233, 387–389
Splitting Rule Interval Criterion property 209
splitting value 176
StatExplore node 10, 50, 51–56, 79, 114, 358
Status Bar 15, 16
Stay Significance Level property 335, 336, 337, 343, 345–347, 372
stepwise selection method
about 343
when target is binary 344–345
when target is continuous 345–347
Stochastic Boosting node 384
stochastic gradient boosting 404
Stop R-Square property, Variable Selection node 73, 74, 121, 157
Sub Tree Method property 198
sub-segments 170
Subtree Assessment Measure property 233
Subtree Method property 377, 387–389, 396–399
SVD (Singular Value Decomposition) 429, 431
SVD Revolution property 451
synthetic variables 246–247
T
Tables to Filter property, Filter node 29
Target Activation Function property 393
target layer 246–247, 270–272
Target Layer Activation Function property 241, 281, 283, 305, 307
Target Layer Combination Function property 241, 270, 281, 283, 305, 307
Target Layer Error Function property 272–273, 281, 283
Target Model property, Variable Selection node 72, 77–78, 78–79, 121–122, 136, 137
target variables, for neural network models 240
targets
See also binary targets
See also continuous targets
See also ordinal targets
compared with samples 8
defining 2–8
maximizing relationship to 96–98
transformations of 98
Targets tab 21
Term Weight property 444
term weighting 440–441, 441–445
term-document matrix 426–427
terminal nodes 130, 170
test data
roles of in development of decision trees 175
testing model performance with 204–205
Test property, Data Partition node 28, 360
Text Cluster node 450, 451–458, 458–461
text files, creating SAS data sets from 433–435
Text Filter node 432, 440–450
Text Filtering node 442
Text Import node 431, 436
text mining, creating data sources for 436
Text Parsing node 431, 436–440, 442, 445–450
Text Topic node 445–450
textual data, predictive modeling with
about 425–426
creating data sources for text mining 436
creating SAS data sets from text files 433–435
dimension reduction 429–431
exercises 461
latent semantic indexing 429–431
quantifying textual data 426–428
retrieving documents from World Wide Web 432–433
Text Cluster node 450, 451–458, 458–461
Text Filter node 432, 440–450
Text Import node 431, 436
Text Parsing node 431, 436–440, 442, 445–450
Text Topic node 445–450
Threshold Significance Level property 184
Time of Kass Adjustment property 184
Time Series node 35–44
%TMFILTER macro 432, 435
Toolbar Shortcut Buttons 15, 16
tools
See nodes
Tools Bar 15
total profit, vs. average profit for comparing tree size 192–193
training, of trees 172
training data
developing trees using 188–189
roles of in development of decision trees 175
training data set 233
Training property, Data Partition node 28, 360
transaction data
converting to time series 35–37
creating data sources for 37–40
transform variables, saving code generated by 163
Transform Variables node
See also variable selection
about 95–101, 117–118, 163, 164
Class Inputs property 376
Hide property 101, 156, 159, 162
Interval Inputs property 95, 96, 98, 155, 159, 162
Merging property 161
Multiple Method property 162
Node ID property 159, 161
Offset Value property 96
in process flow 101, 102, 116
Regression node 359
Reject property 101, 156, 159, 162
testing variables and transformations 45, 46–47, 48
transforming variables 155–157, 158, 159, 160–162
transforming variables with 153–155
Variables property 99
Transformation node 376
transformations
after variable selection 157–159
binning 96
of class inputs 98
for interval inputs 95–98
multiple using Multiple Method property 162
passing more than one for each interval input 159–162
passing two types using Merge node 159–162
simple 96
of targets 98
before variable selection 155–157
of variables 153–163
TRANSPOSE procedure 439
Treat Missing as Level property
Interactive Binning node 84
Regression node 376
trees
about 170
assessing using Average Square Error 194
true positive fraction 258
U
unadjusted probabilities, expected profits using 236
ungrouped variables, compared with categorical variables 165
unordered (nominal) target, regression models with 329–332
Use AOV16 Variables property
Dmine Regression node 313
Variable Selection node 73, 74, 78, 120, 122, 157
Use Group Variables property, Variable Selection node 74–75, 122, 124–125
Use Selection Defaults property 335, 336, 340, 342–343, 386
user-defined networks 307
user-specified architectures 305–307
utility nodes 101–107
V
validation accuracy 193
validation data
pruning trees using 185–187
roles of in development of decision trees 175
validation error 353–354
Validation Error criterion 386
validation misclassification 354
validation profit 175, 190–192
Validation Profit/Loss criterion 355–356
Validation property, Data Partition node 28, 360
variable clustering, using example data set 65–69
Variable Clustering node
about 50, 61–69, 117–118, 163
Include Class Variables property 139–140
Maximum Clusters property 62
Maximum Eigenvalue property 62, 64
Regression node 352
Variable Selection property 139
variable selection using 138–150
variable selection
See also Transform Variables node
about 117–119
binary target with nominal-scaled categorical inputs 135–138
binary target with numeric interval-scaled inputs 129–135
continuous target with nominal-categorical inputs 124–129
continuous target with numeric interval-scaled inputs 119–124
exercises 166–167
transformation after 157–159
transformation before 155–157
using Decision Tree node 150–153
using Variable Clustering node 138–150
Variable Selection node
about 50, 72–79, 163, 164
Hide Rejected Variables property 120, 122
Minimum R-Square property 73, 74, 120–121, 157
regression models 359
Stop R-Square property 73, 74, 121, 157
transforming variables 153–154, 155–157
Variable Selection property
Regression node 377, 389–390
Variable Clustering node 139
variables
assigning to clusters 64–65
categorical 1–2, 165
changing measurement scale of in data sources 164–165
explanatory 241
interval 2
measurement scale of 1–2, 107–109
number of levels of 107–109
selecting for clusters 140–148
synthetic 246–247
transformation of 153–163
types of 107–109
Variables property
about 164
Drop node 80
File Import node 34
Impute node 83
Regression node 95, 123–124, 133–134
Transform Variables node 99
viewing properties 25
variance
of inputs 111
proportion explained by cluster component 65
proportion of explained by principal components 112–113
Variation Proportion property, Variable Clustering node 62
Voting Posterior Probabilities property 408–409
Voting...Average method 408–409
Voting..Proportion method 409
W
weights
estimating in neural network models 247–249
selecting for Neural Network node 261–263
windows
Metadata Advisor Options 19
SAS Enterprise Miner 15–16
World Wide Web, retrieving documents from 432–433
X
XRadial value 307