HealthBeauty360 Retail Analytics

Independent project · synthetic data

What this is

A synthetic-data retail analytics demo: churn, K-Means RFM segmentation, price elasticity, Ridge demand forecasting and PSI/KS drift monitoring, served behind FastAPI in Docker. Every figure is a synthetic-data result.

Metrics

Churn ROC-AUC

0.73

vs recency rule 0.72; majority 0.50

Price elasticity

5 of 7 categories significant, 95% CIs

Segmentation

k=4 by swept silhouette (0.25)

Demand forecast

Beats seasonal-naive on ~77% of SKUs

Stack

DockerFastAPIRidgeK-Means RFMPSI/KS driftMann-KendallGitHub Actions CI

Provenance

Synthetic data throughout; every figure is a synthetic-data result, not a measured commercial outcome. The absolute forecast error is deliberately not published: the only defensible framing is the relative one, that the forecast beats a seasonal-naive baseline on ~77% of SKUs.

What I would not claim

Isolation Forest or anomaly detection
Prophet, XGBoost, SHAP or an ensemble
Great Expectations, a feature store, dbt, BigQuery or Cloud Run
production-grade
the absolute forecast error