🧠 Comparing Clustering Algorithms by Customer Segmentation

This project applies unsupervised machine learning techniques to perform customer segmentation on a marketing dataset. It uses the BIRCH, K-Means and DBSCAN clustering algorithms to group customers based on their demographic and behavioral features, with a focus on interpretability, performance, and scalability.

📁 Dataset Description

The dataset is a cleaned and scaled version of the Customer Personality Analysis dataset available on Kaggle. It contains features such as:

• Demographic Features: Age, Income, Family Size
• Behavioral Features: Total Spend, Recency, Tenure, Campaign Responses, Purchases via different channels (Web, Catalog, Store)
• Target: no labeled output

🧪 Algorithms Used

• ✅ BIRCH Clustering

• ✅ KMeans Clustering

• ✅ DBSCAN

• ✅ PCA for dimensionality reduction (visualization)

🔍 Project Goals

• Segment customers into meaningful groups

• Perform 3D visualization of clusters using PCA

• Extract cluster-wise insights based on statistical summaries

• Generate automatic customer personas per cluster

• Make the code modular and reusable for other models

📦 Project Structure

customer-segmentation

│

├── data/

│ └── processed_customers.csv # Cleaned and scaled dataset

│

├── models/

│ ├── birch_clustering.ipynb # BIRCH analysis

│ ├── kmeans_clustering.ipynb # KMeans analysis

│ ├── dbscan_clustering.ipynb # DBSCAN analysis

│ └── persona_generator.py # assign persona to each cluster

│

├── notebooks

│ ├── eda.ipynb # Exploratory data analysis

│ ├── data_preprocessing.ipynb

│ ├── model_comparison.ipynb

│

├── processed

│ ├── birch.csv # dataset containing cluster labels by BIRCH Algorithm

│ ├── kmeans.csv # dataset containing cluster labels by K-Means Algorithm

│ ├── dbcsan.csv # dataset containing cluster labels by DBSCAN Algorithm

│ ├── processed_customers.csv # dataset after performing data manipulation

│ ├── processed_customers_raw.csv #

│ └── processed_customers_unscaled.csv

│

├── README.md # This file

└── requirements.txt # Dependencies

📊 Cluster Analysis Example (BIRCH)

Each cluster is analyzed using range estimates: (mean ± std) for features like:

• Income: Customer’s income
• Recency: Days since last purchase
• Total Spend: Combined spend across all channels
• Customer Tenure: Days since becoming a customer
• Web, Catalog, and Store Purchases
• Campaign Responses, Complaints, Family Size, Age

Insights are used to label each cluster with customer personas, like:

• 🎯 "High-Value, Multi-Channel Spenders"
• 🧍 "Low-Engagement, Budget Shoppers"
• 📦 "Catalog Loyalists with High Income"

📈 3D Visualization

Clustering results are visualized using 3D PCA plots to interpret how customer segments are distributed in reduced dimensions

🧠 Future Enhancements

• Implement CURE hierarchical clustering
• Add cluster evaluation metrics (Silhouette, Davies-Bouldin)

🚀 Getting Started

1. Clone the repository:

   git clone https://github.com/shubhro2002/Comparing-Clustering-Algorithms-by-Customer-Segmentation.git
   

2. Install required dependencies:

   pip install -r requirements.txt
   

3. Run the Jupyter notebooks or Python scripts to start using.