Retail Customer Retention Tutorial

Notebooks

Click on a notebook to view the tutorial. Code cells are collapsed by default - click "Show/Hide Code" to expand.

Production Validation Highlights

The scoring validation notebook (11) evaluates model performance on a point-in-time holdout set and analyzes distribution drift.

Key Findings:

XGBoost achieved best validation ROC-AUC (0.9726)
Logistic Regression: ROC-AUC 0.9648, most robust to distribution shift
SHAP analysis reveals top predictive features
Holdout validation confirms model generalization

About This Tutorial

This tutorial demonstrates the Customer Retention ML Framework using a synthetic retail dataset (30,801 customers). Each notebook walks through a stage of the ML pipeline, from data exploration to production validation.

Pipeline stages covered:

Automatic data profiling and type detection
Temporal scenario detection and point-in-time splits
Data quality assessment and source integrity checks
Feature engineering and capacity analysis
Model training: Logistic Regression, Random Forest, XGBoost
SHAP interpretability and feature importance
Pipeline code generation (local and Databricks)
Distribution drift analysis and production validation