Logistic Regression

1
Logistic Regression

• Demo dmdata2 and dmdata3
• Bankloan
• Assignment subscribe_training and
  subscribe_validate

2
Purpose

• Steps to follow with the logistic regression
  procedure
• This follows the initial propensity to buy from
  the direct marketing modules

3
Data Preparation 1

1. Compare the two files that must have the same
   measure indicator
2. Analyse response frequency from the training file
3. Determine file size from split
4. Determine break-even point (FC/GM) available from
   the SM Template.

4
Data Preparation Trainining 2

• In the Subscribe_training, response2 is coded 2
  non-respondent and 1 respondent. You must
  recode directly or in another variable so that
  non-respondent 0 and respondent 1. Change
  variable labels.
• Transform ? Random number generator ? Set
  starting value ? Fix value 9-digit student id.
• Split file for training and testingCompute
  select rv.bernouilli (0.70)
• Print your log file into SM

5
Logistic Regression Training 1

• Analyse ? Regression ? Binary Regression
• Enter dependent and independent variables
• Method Forward LR
• Enter select into Selection variable

6
Logistic Regression Training 2

• Categorical ? Enter non-scale variables into the
  Categorical Covariates
• Change Contrast (indicator, First reference) and
  click Change

7
Logistic Regression Training 3

• Save
• Predicted Values Probabilities
• Export XML Model (Name model)
• Option
• Display at last step
• Classification cutoff
• Run

8
Logistic Regression Training 3

1. Paste Regression Log syntax into SM
2. Enter classification matrix into SM
3. Create Deciles from Predicted Value using Visual
   Binning
4. Compare Means ? Means ? Analyse Resp2 by Deciles
5. Enter Mean response by deciles into SM

9
Logistic Regression Validation 1

1. Score Validation file from the XML saved earlier
2. Create Deciles from the predicted values using
   Visual Binning.
3. Paste your Visual Binning log syntax into SM
4. Compare Means ? Means ? Analyse Predicted values
   by Deciles
5. Enter Mean of predicted values by deciles into SM

10
Propensity to buy and Logistic Regression

• You are responsible to do the appropriate
  marketing metrics.