target marketing for donation soliciting
TRANSCRIPT
Target Marketing For Donation Soliciting By: Sai Praneeth
Project Methodology1. Problem Definition2. Data Cleansing 3. Data Visualization4. Model Building/Training5. Model Comparison/ Validation 6. Model Scoring
Problem Definition Current State
• A veterans organization seeks to better target its solicitation for donation.
• So that less money is spent on solicitation efforts and more money is available for charity.
Future State
• Develop a predictive model to identify potential donors.
• Reduce cost and increase donations for the organizations.
Note: Soliciting involves sending greeting cards, emails, phone calls etc. along with a request for donation
Data Overview
Target Description
1 Donated
0 No donation
Response Variable
Predictors DescriptionDem Age Age
Dem Gender Gender
Dem Home owner
Home owner
Dem Income Median Income
Gift Avg 36 Gift Amount Average 36 months
Gift Avg all Gift Amount Average all months
Gift time last Time since last gift
PromCnt12 Promotion Count card 12 months
StatusCat96NK
Status Category 96nk
Predictors
Data CleaningChecking for Missing ValuesRare event over Sampling Check for data entry errorsCheck for DuplicatesCheck for Outliers
Data Visualization
Model Building As our model has binary response there are 2 possible predictive models.
1. Decision Trees
2. Logistic Regression
Decision TreesStep 1: Data Partition
o Training (50%)
Variable Numeric Value
Frequency Count
Percent
Target_B 0 2422 50.01Target_B 1 2421 49.98o Validation (50%)
Variable Numeric Value
Frequency Count
Percent
Target_B 0 2421 49.98Target_B 1 2422 50.01
Step 2: Model Construction
Maximal Decision Tree
Step 3: Model Assessment
Note: Cleary there is a problem of overfitting.
Step 4: Model Optimization
• The ASE is for the validation data set is least at 5 leaves, which is the optimal tree.
Step 5: Optimal Decision tree
Decision Rule: People who have made donations 3 or more times in the past 36 months & the donated amount was less then $ 7.5 , have 66% chance of making a donation now.
Logistic RegressionStep 1: Input Selection
o Forward Selection
o Backward Selection
o Stepwise Selection In this case stepwise selection with P-Value of 0.05 was chosen.
Step 2: Model Assessment
• From the graph it is evident that the Misclassification rate for the validation data set is least at step 3, which is the optimal model.
Step 3: Optimal Model
Based on the P-Value the significant variables are:
o Dem Median Home Value
o Giftcnt36
o GiftTimeLast
Odds Ratio Estimates:
Effect Point EstimateDemMedHomeValue
1.00
Giftcnt36 1.121GiftTimeLast 0.965
Step 4: Model Interpretation
o For Giftcnt36 odds ratio is 1.121; this means that for each additional donation in the 36 months, the odds of donation increases by 12.1%.
o For GiftTimeLast the odds ratio is 0.965. That is if the time since last gift increases by 1 month, the odds of donation decrease by 3.5%.
o For DemMedianHomeValue the odds ratio is 1. which indicates that median home value does not significantly impact donations.The conclusion obtained from logistic regression model is similar to that obtained from decision tree.
Model ComparisonSelectedModel
Model Type
ValidationMisclassification
Validation ASE
Y Decision Tree
0.42804 0.2433
N Regression 0.4391 0.2442
Adjustment for oversampling
SelectedModel
Model Type
ValidationMisclassification
Validation ASE
N Regression 0.5001 0.2442
Y Decision Tree
0.5001 0.2432
Adjusted Fit-Statistics
ROC Chart
Lift Plot
Model Scoringo Once the best model is selected, implement the model on a
scoring data set.
o Then export the scored data set to a .csv or SAS file for implementation.
Miscellaneouso Check for Missing Values
o Check for Outliers
o Over Sampling
o Data entry errors
Thank You
Any questions?