Online Retail II Customer Segmentation

Approach

• Clean transactions: remove missing customer IDs, cancellations, and non-positive values.
• Create customer-level features: Recency, Frequency, Monetary.
• Apply log transform to Frequency and Monetary to reduce skew.
• Scale features, test multiple k values, and select by silhouette + interpretability.
• Assign business-friendly segment names and recommendations.

Model Selection

KMeans with k=3 achieved the strongest baseline separation.

• KMeans silhouette: 0.4117
• GMM used as comparison model (check notebook output for latest score)

Why k=3: balanced cluster quality and easy stakeholder interpretation.

Segment Definitions

Results Visualization

Customers per Segment

Avg Revenue by Segment

PCA Projection

Two-dimensional projection used for communication, not for model training.

Deliverables and Deployment

• Notebook: notebooks/01_customer_segmentation_starter.ipynb
• Data output: data/processed/customer_segments.csv
• Report: reports/customer_segmentation_report.html
• Deploy site: bash scripts/deploy_report.sh then open docs/index.html

Project status: baseline complete and presentation-ready.