Business Context
A financial services company wants to predict which customers are likely to respond to a new loan offer. The goal is to build a reliable classification model using customer demographics and historical data, partition the data for validation, and save the final model for future scoring of new customer lists.
Data Preparation
Create a dataset simulating customer profiles and their response to a previous campaign. Predictors include Age, Income, number of existing products, and home ownership status. The target 'Responded' is binary (1/0).
Copied!
| 1 | DATA customer_offers; |
| 2 | call streaminit(2024); |
| 3 | DO CustomerID = 1 to 20000; |
| 4 | Age = 22 + floor(rand('Uniform') * 48); |
| 5 | Income = 45000 + floor(rand('Uniform') * 150000); |
| 6 | ExistingProducts = 1 + floor(rand('Uniform') * 5); |
| 7 | IF rand('Uniform') < 0.6 THEN OwnsHome = 'Yes'; |
| 8 | ELSE OwnsHome = 'No'; |
| 9 | logit_p = -2.5 + (Age / 20) - (Income / 90000) + (ExistingProducts * 0.3) - (ifc(OwnsHome='Yes', 0.5, 0)); |
| 10 | p = 1 / (1 + exp(-logit_p)); |
| 11 | Responded = rand('Bernoulli', p); |
| 12 | OUTPUT; |
| 13 | END; |
| 14 | RUN; |
| 15 | |
| 16 | DATA casuser.customer_offers_train; |
| 17 | SET customer_offers; |
| 18 | RUN; |
Étapes de réalisation
1
Load the customer data into CAS and verify the table structure.
Copied!
| 1 | PROC CAS; |
| 2 | TABLE.tableInfo TABLE={caslib='CASUSER', name='customer_offers_train'}; |
| 3 | RUN; |
| 4 | QUIT; |
2
Run bart.bartProbit, partitioning the data (75% train, 25% test), specifying 150 trees, and saving the model.
Copied!
| 1 | PROC CAS; |
| 2 | bart.bartProbit TABLE={name='customer_offers_train'}, |
| 3 | target='Responded', |
| 4 | inputs={'Age', 'Income', 'ExistingProducts'}, |
| 5 | nominals={'OwnsHome'}, |
| 6 | partByFrac={test=0.25, seed=1985}, |
| 7 | nTree=150, |
| 8 | nBI=200, |
| 9 | nMC=1500, |
| 10 | seed=1234, |
| 11 | OUTPUT={casOut={name='customer_response_preds', replace=true}, pred='P_Responded'}, |
| 12 | store={name='bart_loan_model', replace=true}; |
| 13 | RUN; |
| 14 | QUIT; |
3
Check the created output and model store tables.
Copied!
| 1 | PROC CAS; |
| 2 | TABLE.tableInfo TABLE={name='customer_response_preds'}; |
| 3 | TABLE.tableInfo TABLE={name='bart_loan_model'}; |
| 4 | RUN; |
| 5 | QUIT; |
Expected Result
The action should successfully train a model and produce fit statistics for both training and testing partitions. A CAS table 'customer_response_preds' will be created containing the predicted probabilities for each customer. A model store table 'bart_loan_model' will also be created in the active caslib, ready for use with the bartScore action.