Customer Segmentation with KMeans

This project segments mall customers using KMeans clustering on annual income and spending score. It includes exploratory analysis, model selection visuals, clustering, and evaluation metrics.

Dataset

Mall_Customers.csv includes the following columns:

• CustomerID
• Gender
• Age
• Annual Income (k$)
• Spending Score (1-100)

Methodology

• Exploratory analysis of feature distributions
• Feature scaling with StandardScaler
• Elbow method and silhouette score for cluster selection
• KMeans clustering (k=5)
• Cluster profiling (mean income/spending and count)

Outputs

All outputs are saved in reports/:

• eda_income_distribution.png
• eda_spending_distribution.png
• eda_income_vs_spending.png
• elbow_method.png
• silhouette_by_k.png
• davies_bouldin_by_k.png
• calinski_harabasz_by_k.png
• customer_segments.png
• cluster_profile.csv
• metrics.csv
• metrics_by_k.csv
• segmented_customers.csv

How to Run

python kmeans_customer_segmentation.py

Results Snapshot

• Evaluation metrics are saved to reports/metrics.csv
• Cluster profiles are saved to reports/cluster_profile.csv

Interpreting the Metrics

• Silhouette Score: ranges from -1 to 1. Higher is better.
• Davies-Bouldin Index: lower is better.
• Calinski-Harabasz Index: higher is better.

To justify k=5, compare metrics across k in reports/metrics_by_k.csv alongside the elbow and silhouette plots.

Best K Summary

Use these heuristics together:

• Pick a k at the elbow where WCSS reduction slows.
• Prefer higher silhouette scores.
• Prefer lower Davies-Bouldin scores.
• Prefer higher Calinski-Harabasz scores.

If all three agree around k=5, it is a strong, defensible choice.

Notes for Analysts

This project demonstrates:

• Proper preprocessing and scaling
• Model selection with multiple diagnostics
• Clear, reproducible outputs
• Interpretable cluster summaries for stakeholders