Complete Customer Analytics with Online Retail II Dataset

📝 Project Summary

This project provides a comprehensive, end-to-end customer analytics solution using the Online Retail II dataset. It transforms over one million raw transactional records into actionable business insights. The core of this project involves:

• RFM Segmentation: Identifying distinct customer segments such as "Champions" and "At-Risk" by analyzing their Recency, Frequency, and Monetary value to enable targeted marketing strategies.
• Predictive Churn Modeling: Building and deploying an XGBoost model to proactively identify customers with a high probability of churning, and analyzing the key factors that drive this behavior.
• Customer Lifetime Value (CLV) Forecasting: Implementing probabilistic models (BG/NBD and Gamma-Gamma) to predict the future purchasing behavior and monetary value of customers, guiding long-term strategic investments.
• Statistical A/B Testing: Designing and executing a simulated A/B test to statistically measure the impact of a discount campaign on customer spending, using the Mann-Whitney U test for significance.

🎯 Project Objectives

The main business objectives of this project are:

• Identify the most valuable customer segments and their common characteristics.
• Pinpoint customers at high risk of churn and understand the factors influencing this risk.
• Estimate the future value (CLV) of customers to guide marketing budget allocation.
• Determine the effectiveness of a marketing campaign on increasing customer spending.

🛠️ Tech Stack

📂 Project Structure

• data/: Contains raw, intermediate, and final datasets.
• images/: Stores visual assets used in the README.
• notebooks/: Houses the Jupyter Notebooks for each analysis stage.
  • 01_Data_Preprocessing_EDA.ipynb
  • 02_RFM_and_Segmentation.ipynb
  • 03_Churn_Prediction.ipynb
  • 04_CLV_Analysis.ipynb
  • 05_AB_Testing.ipynb
• pyproject.toml, uv.lock: Files for modern dependency and environment management.
• README.md: This project overview.

📊 Data Dictionary

The dataset contains transactional data from a UK-based online retail company. Key fields used in the analysis include:

Feature	Description	Type
Invoice	A 6-digit nominal, unique for each transaction. 'C' indicates a cancellation.	object
StockCode	A unique code assigned to each distinct product.	object
Description	The name of the product.	object
Quantity	The number of units of a product sold in a transaction.	int64
InvoiceDate	The date and time when the transaction occurred.	object
Price	The unit price of the product in British Pounds (£).	float64
Customer ID	A 5-digit unique identifier for each customer.	float64
Country	The country where the customer resides.	object

🔄 Analysis Workflow & Notebook Details

This project follows a sequential workflow, with each step detailed in its respective Jupyter Notebook.

1. Data Cleaning and EDA (notebooks/01_Data_Preprocessing_EDA.ipynb)

   • Cleans and preprocesses over 1 million records, handling missing values, duplicates, and cancellations.
   • Performs feature engineering to create analysis-ready features.
   • Applies IsolationForest for sophisticated outlier detection.
   • Conducts comprehensive exploratory data analysis (EDA) to uncover trends in sales over time, by country, and by product.

2. RFM Analysis & Segmentation (notebooks/02_RFM_and_Segmentation.ipynb)

   • Calculates Recency, Frequency, and Monetary (RFM) metrics for each customer.
   • Segments customers into actionable groups like "Best Customers," "At-Risk," and "Promising Customers."
   • Visualizes segment distributions and characteristics using 2D and 3D plots.

3. Churn Prediction (notebooks/03_Churn_Prediction.ipynb)

   • Defines churn based on a 90-day inactivity window.
   • Builds and compares Logistic Regression, Random Forest, and XGBoost models.
   • Evaluates models using ROC-AUC, F1-score, and confusion matrices to identify the best-performing algorithm.
   • Visualizes feature importance to understand key drivers of customer churn.

4. CLV Analysis (notebooks/04_CLV_Analysis.ipynb)

   • Implements the probabilistic BG/NBD model to predict future transaction frequency.
   • Uses the Gamma-Gamma model to estimate the average monetary value of transactions.
   • Develops a hybrid XGBoost model using BG/NBD features for enhanced CLV prediction.
   • Compares CLV results with RFM segments for validation.
   • Visualizes CLV distribution and identifies key customer groups.

5. A/B Testing (notebooks/05_AB_Testing.ipynb)

   • Simulates an A/B test scenario to measure the impact of a discount campaign.
   • Applies appropriate statistical tests (Shapiro-Wilk for normality, Mann-Whitney U for comparison) to determine statistical significance.
   • Provides clear visualizations and reporting of the test results.

🚀 Setup and Execution

Environment Setup (Windows PowerShell)

1. Ensure Python 3.12+ is installed.
2. Clone the repository:

   git clone https://github.com/mertafacan/Complete-Customer-Analytics-for-E-Commerce.git
   cd Complete-Customer-Analytics-for-E-Commerce

3. (Recommended) Create and activate a virtual environment:

   python -m venv .venv
   .\.venv\Scripts\Activate.ps1

4. Install the required dependencies using a modern package manager. uv is recommended for its speed.
   • Using uv (Recommended):

     pip install -U uv  # Install uv if you haven't already
     uv pip install -e .

   • Using pip:

     pip install -U pip
     pip install -e .

5. Launch Jupyter Notebook or JupyterLab and navigate to the notebooks/ directory to explore the analyses. (Optional)

   jupyter notebook

📬 Contact

Mert Afacan – https://www.linkedin.com/in/mert-afacan/ – mert0afacan@gmail.com