This project predicts which customers will respond to a promotion campaign using logistic regression and 10-fold cross validation. The result is read through lift and a gains curve so managers can decide what share of top-ranked customers is worth contacting.
Business Goal
Choose who to contact so the campaign budget yields the most responses. Application area: Marketing. Project level: beginner.
RapidMiner Tools and Operators
The process is built with these operators:
- Read CSV
- Remap Binominals
- Nominal to Numerical
- Normalize
- Cross Validation
- Logistic Regression
- Apply Model
- Performance (Binominal Classification)
- Performance to Data
- Write CSV
What You Will Learn
- Understand logistic regression as a probability model for marketing decisions
- Read lift and a gains curve to size a campaign target
- Estimate campaign profit from model scores and cost assumptions
Working Steps
- Read the campaign data — Read CSV loads data_kampanye.csv (4,000 customers). id_pelanggan becomes the id and respons the binominal label (yes/no).
- Define the positive class — Remap Binominals makes "ya" (responded) the positive class.
- Turn categorical attributes into numbers — Nominal to Numerical (dummy coding) turns preferred channel and membership status into 0/1 columns usable by logistic regression.
- Rescale the attributes — Normalize (Z-transformation) makes model weights comparable across attributes, for example email open rate versus days since last purchase.
- Train Logistic Regression inside Cross Validation — Cross Validation with 10 folds (stratified). The Training part holds Logistic Regression with a dot (linear) kernel and C = 30.
- Test the model — Apply Model produces confidence(ya) for each test customer; Performance (Binominal Classification) computes accuracy, AUC, precision and recall.
- Save each customer's score — Write CSV saves the performance measures and per-customer predictions. The final model goes to a result port so the attribute weights can be read in the Description tab.



