customer_segmentation

Customer segmentation using RFM analysis, UMAP, and clustering

Online Gift Store Customers Segmentation

📌 Project Overview

This project focuses on customer segmentation for an online gift store using RFM analysis and clustering techniques. The goal is to identify distinct customer groups based on purchasing behavior to enhance targeted marketing strategies and improve customer relationship management.

📂 Dataset

The dataset is sourced from The UCI Machine Learning Repository and contains transactional records from an online gift store operating between 01/12/2010 and 09/12/2011. The company sells unique gifts, with many of its customers being wholesalers.

Features:

• InvoiceNo – Unique transaction ID (prefix 'C' indicates a canceled transaction).
• StockCode – Unique product ID.
• Description – Product name.
• Quantity – Number of units purchased.
• InvoiceDate – Date and time of transaction.
• UnitPrice – Price per unit in GBP (£).
• CustomerID – Unique customer identifier.
• Country – Customer's country.

🔧 Data Preprocessing

1. Handling Missing Values
   • Removed transactions with missing CustomerID and Description.
2. Duplicate Removal
   • Identified and removed duplicate entries.
3. Processing Returns & Special Transactions
   • Created QuantityCancelled to track returned items.
   • Removed transactions containing special product codes (e.g., postage fees, bank charges).
4. Filtering Outliers
   • Removed transactions with zero or negative unit prices.
5. Feature Engineering
   • Calculated TotalPrice (UnitPrice * (Quantity - QuantityCancelled)).
   • Extracted Purchase Time Features (Month, Weekday, Hour).
   • Created a IsWholesale indicator (bulk orders).
   • Created Unique Items Per Order feature.

📊 Exploratory Data Analysis (EDA)

Key insights extracted:

• Top buying countries (most orders and revenue contributors).
• Seasonality in sales (peak months, days, and hours of transactions).
• Most valuable customers based on spending.
• Frequent return patterns and affected products.
• Average Order Value (AOV) distribution.

🔍 Customer Segmentation (RFM Analysis)

To cluster customers, we calculated the Recency, Frequency, and Monetary (RFM) scores:

• Recency – Days since last purchase.
• Frequency – Total number of unique purchases.
• Monetary – Total spending amount.

🧠 Clustering Methods & Model Selection

Dimensionality Reduction:

• PCA – Used to reduce high-dimensional data to 2D space.
• t-SNE – Used to reduce high-dimensional data to 2D space.

• UMAP – Used to reduce high-dimensional data to 2D space (showed the best results).

Clustering Algorithms Tested:

• K-Means (Best Model: K-Means with 5 Clusters, Silhouette Score: 0.6029)
• Agglomerative Clustering (Silhouette Score: 0.5897)
• DBSCAN (Density-Based Clustering)
• Gaussian Mixture Model (GMM) 

Handling Outliers:

• Used Isolation Forest to remove extreme outliers before clustering.

Visualizations:

• Top 10 Countries by Revenue [View Chart] (https://wsiqz.github.io/customer_segmentation/top_10_countries_by_revenue.html)

• Monthly Revenue (see Project)

• Daily Orders over Time (see Project)

• Sales Trend by Day of the Week [View Chart] (https://wsiqz.github.io/customer_segmentation/sales_trend_by_day_of_the_week.html)

• Orders by Hour (see Project)

• Top 10 Customers by Total Spending [View Chart] (https://wsiqz.github.io/customer_segmentation/top_10_customers_by_total_spending.html)

• Distribution of AOV [View Chart] (https://wsiqz.github.io/customer_segmentation/distribution_of_average_order_value.html)

• Top 10 Returned Products [View Chart] (https://wsiqz.github.io/customer_segmentation/top_10_returned_products.html)

• RFM Boxplots [View Chart] (https://wsiqz.github.io/customer_segmentation/rfm_boxplots.html) (https://wsiqz.github.io/customer_segmentation/rfm_boxplots_filtered.html)

• 2D RFM Scatter Plot [View Chart] (https://wsiqz.github.io/customer_segmentation/rfm_scatter_plot.html)

• 2D RFM PCA Scatter Plot [View Chart] (https://wsiqz.github.io/customer_segmentation/rfm_pca_scatter_plot.html)

• 2D RFM t-SNE Scatter Plot [View Chart] (https://wsiqz.github.io/customer_segmentation/rfm_tsne_scatter_plot.html)

• 2D RFM t-SNE Clusters Scatter Plot [View Chart] (https://wsiqz.github.io/customer_segmentation/rfm_tsne_clusters.html)

• 2D RFM UMAP Scatter Plot [View Chart] (https://wsiqz.github.io/customer_segmentation/rfm_umap_scatter_plot.html)

• 2D RFM UMAP Clusters Scatter Plot [View Chart] (https://wsiqz.github.io/customer_segmentation/rfm_umap_clusters.html)

• 2D RFM UMAP (no outliers) Clusters Scatter Plot [View Chart] (https://wsiqz.github.io/customer_segmentation/rfm_umap_clusters_clean.html)

• 3D Final Clusters Scatter Plot [View Chart] (https://wsiqz.github.io/customer_segmentation/rfm_3d_clusters_clean.html)

• Cluster Profile Radar Chart [View Chart] (https://wsiqz.github.io/customer_segmentation/cluster_profile.html)

🔍 Key Findings (Customer Segments)

1. Loyal High-Spenders – Frequently purchase and spend large amounts.
2. Occasional Buyers – Infrequent orders but moderate spending.
3. New Customers – Recently made first purchases.
4. High-Value Inactive Customers – Previously spent a lot but stopped buying.
5. Discount Hunters – Small but frequent purchases, possibly during sales.
6. Churned Customers – Haven’t purchased in a long time with low spending history.

🚀 How to Run the Project

1. Clone the repository:

   git clone https://github.com/wsiqz/customer_segmentation.git
   cd customer_segmentation

2. Install dependencies:

   pip install -r requirements.txt

3. Run the Jupyter Notebook:

   jupyter notebook PROJECT-6._Online_gift_store_customers_segmentation.ipynb

🛠 Technologies Used

• Python (Pandas, NumPy, Scikit-learn, Plotly, UMAP)
• Machine Learning (Clustering, Outlier Detection, Dimensionality Reduction)
• Jupyter Notebook (For interactive analysis)

📁 Repository Structure

📂 customers_segmentation
│-- 📄 PROJECT-6._Online_gift_store_customers_segmentation.ipynb
│-- 📄 README.md
│-- 📄 requirements.txt
│-- 📂 data/ (Cleaned transactions dataset, Original dataset)
|-- 📂 images/ (Plots(png))
|-- 📂 docs/ (Plots(html))

🎯 Future Improvements

• Implement Deep Learning-based clustering for better accuracy.
• Apply time-series forecasting for predicting future customer behavior.
• Develop automated dashboards for real-time insights.

📜 License

This project is open-source and available under the MIT License.

---

Developed by Mariya Kostyrya – Customer Segmentation Using Machine Learning 📊🚀