Online Retail — K-Means Customer Segmentation

---

Goal: Use unsupervised learning (K-Means) on RFM features to uncover actionable customer segments from the UCI Online Retail dataset. Business Impact: Customer segmentation drives personalized marketing, improves ROI, and enhances customer retention & engagement.

---

🚀 Executive Summary

This notebook demonstrates a complete end-to-end ML pipeline that:

• Cleans & standardizes messy retail transactions
• Constructs RFM (Recency, Frequency, Monetary) features
• Scales features & applies K-Means clustering
• Determines optimal k via Elbow & Silhouette methods
• Produces 4 actionable segments + 3 outlier micro-segments
• Visualizes clusters for business storytelling

📊 Key Outcomes:

• Raw rows: 525,461 → Cleaned: 406,309 (77.32% retained)
• Unique customers: 4,285
• Clustered customers (after outlier removal): 3,809
• Optimal clusters (k): 4
• Actionable Segments: Retain, Reward, Re-Engage, Nurture + Outlier cohorts (Pamper, Upsell, Delight)

---

📂 Dataset

• Source: UCI ML Repository — Online Retail II
• Period: 2009-12-01 → 2010-12-09
• Countries: 40 (UK dominates with 485,852 rows)
• Schema (8 cols): Invoice, StockCode, Description, Quantity, InvoiceDate, Price, Customer ID, Country

---

🧹 Data Cleaning

Steps applied:

1. Invoice filtering → keep only 6-digit numeric invoices (^\d{6}$).
2. StockCode filtering → exclude admin/test codes (keep only SKU-like).
3. Drop rows with missing Customer IDs.
4. Remove zero-priced lines (28 rows removed).
5. Final cleaned dataset: 406,309 rows (77.32% of raw).

📌 Distribution Check:

📌 With Outliers Highlighted:

📌 After Major Outlier Seperated:

---

📊 RFM Feature Engineering

Features per customer:

• Recency (R): Days since last purchase
• Frequency (F): Number of invoices
• Monetary (M): Total spend

Summary (non-outliers, 3,809 customers):

• Recency: mean = 97d, median = 58d
• Frequency: mean = 2.86, median = 2
• Monetary: mean = 885.5, median = 588.1

---

🤖 Modeling

• Algorithm: KMeans(n_clusters=4, random_state=42, max_iter=1000)
• Scaling: StandardScaler on [R, F, M]
• Model Selection:

📌 Elbow + Silhouette Diagnostic:

Optimal K = 4

---

🧩 Segmentation Results

🔑 Core Clusters

1. Cluster 0 — Retain (High value, recent buyers)

   • Playbook: VIP care, exclusive perks, proactive engagement.

2. Cluster 1 — Re-Engage (Low frequency & spend, lapsed)

   • Playbook: Win-back offers, personalized reactivation emails.

3. Cluster 2 — Nurture (Lowest activity, often new)

   • Playbook: Onboarding, education, low-friction deals.

4. Cluster 3 — Reward (Loyal, consistent buyers)

   • Playbook: Loyalty rewards, referral incentives, bundles.

📌 Cluster Visualization (Raw):

📌 Cluster Visualization (Scaled):

📌 KMeans Cluster Separation:

📌 Violin Plots (Segment Profiles):

---

🎯 Outlier Micro-Segments

• Cluster −1 — Pamper → High spenders → bespoke offers, concierge.
• Cluster −2 — Upsell → Frequent shoppers → subscriptions, bundles.
• Cluster −3 — Delight → Elite (high R + F + M) → premium tiers, surprise perks.

---

💼 Business Applications

• Retention & Loyalty → Retain & Reward cohorts
• Reactivation Campaigns → Target Re-Engage group
• Acquisition Funnel → Nurture new/dormant customers
• Premium Strategy → Outlier groups (Pamper, Upsell, Delight)

---

⚙️ Tech Stack

• Python 3.10+, Jupyter Notebook
• pandas → data manipulation
• scikit-learn → KMeans, scaling, silhouette
• matplotlib & seaborn → visualization
• openpyxl → Excel I/O

---

🌟 Highlights

✔️ Built a scalable, reproducible ML pipeline from raw retail logs

✔️ Applied RFM analysis + K-Means for customer segmentation

✔️ Delivered business-ready insights & playbooks tied to ROI levers

✔️ Integrated robust data cleaning, outlier handling, and model diagnostics

✔️ Produced clear visualizations & segment storytelling for stakeholders