🛒 Customer Segmentation Using Clustering

📘 Project Overview

This project performs customer segmentation for a supermarket chain to improve membership card conversion rates using unsupervised machine learning.
By identifying distinct customer groups based on demographics and spending behavior, marketing teams can craft personalized offers and improve engagement.

---

🎯 Problem Statement

Supermarkets face challenges in increasing membership card sign-ups because marketing efforts target all customers uniformly.
The goal is to segment customers into distinct groups to enable:

• Personalized marketing
• Better understanding of customer behavior
• Improved conversion and retention rates

---

📂 Dataset Information

• Source: Provided dataset (/data)
• Records: 200 customers
• Features:
  • Age
  • Gender (encoded)
  • Annual Income
  • Spending Score

Preprocessing Steps:

• Handled missing values
• Encoded categorical variables
• Applied feature scaling (StandardScaler)
• Split data into train and test sets

---

⚙️ Workflow Summary

1. Exploratory Data Analysis (EDA)
   • Visualized feature distributions and correlations
   • Generated heatmaps and pairplots
2. Feature Engineering
   • Standardized numerical features
3. Model Training & Evaluation
   • Trained multiple clustering algorithms:
     • KMeans
     • Gaussian Mixture Model (GMM)
     • Agglomerative Clustering
     • DBSCAN
4. Model Evaluation Metrics
   • Silhouette Score
   • Davies–Bouldin Index
   • Calinski–Harabasz Score
5. Cluster Profiling
   • Analyzed customer characteristics per cluster
   • Exported profiles to CSV

---

🤖 Model Comparison

Model	Train Silhouette	Test Silhouette	After Tuning	Observations
KMeans	0.3627	0.4056	0.4280	✅ Stable and generalizes well
GMM	0.4224	0.3257	0.4285	⚠ Slight overfitting
Agglomerative	0.7229	—	0.7355	⚠ Overfitted to training data
DBSCAN	0.5161	—	0.5506	⚠ Sensitive to parameters

---

🧠 Final Model Recommendation — KMeans

Why KMeans?

• Consistent performance on train and test sets
• Tuned Silhouette Score: 0.4280
• Simple, interpretable, and scalable
• Capable of predicting new customer segments easily

Business Impact:

• Enables targeted marketing for high-value clusters
• Boosts membership card conversion
• Supports automated segmentation for future data

---

📊 Key Visuals

Stored in out/images/:

Visualization	Description
corr_heatmap.png	Feature correlations
eda_01.png – eda_03.png	Exploratory analysis
optimal_k_sil.png	Optimal cluster (KMeans)
model_comparision.png	Model comparison summary
cluster_before_tuning.png, cluster_after_tuning.png	Clustering improvement

Cluster summaries available in:
out/files/cluster_profiles/

---

🧩 Saved Models & Artifacts

Located in out/models/:

File	Description
KMeans_model.pkl	Final selected model
GMM_model.pkl	Gaussian Mixture Model
Agglomerative_model.pkl	Hierarchical model
DBSCAN_model.pkl	Density-based model
scaler.pkl	Feature scaler

---

🧪 How to Run

1. Open the notebook:

   retrain.ipynb

2. Execute all cells to:
   • Load and preprocess data
   • Train and evaluate clustering models
   • Save outputs (images, CSVs, and models)

Outputs will automatically be saved in /out/files/ and /out/models/.

---

🗂️ Project Structure

project/
│
├── data/                          # Raw dataset
├── out/
│   ├── files/
│   │   └── cluster_profiles/      # Cluster CSV reports
│   ├── images/                    # Visualizations
│   └── models/                    # Trained models and scaler
│
├── retrain.ipynb                  # Main notebook
├── README.md                      # Project documentation
└── .gitattributes                 # Git configuration

---

🚀 Future Improvements

• Apply PCA for dimensionality reduction before clustering
• Add behavioral metrics (purchase frequency, product categories)
• Deploy as an interactive dashboard or API for real-time segmentation

---

👤 Author

Name: Isula Dissanayake
Date: November 2025
Project: Customer Segmentation using Machine Learning

---

⭐ Final Takeaway:
KMeans provides the most interpretable, stable, and business-ready segmentation model for optimizing membership marketing strategies.