📊 Customer Churn Prediction System

A complete end-to-end churn prediction pipeline combining SQL, Python (ML), and Power BI to identify at-risk telecom customers and visualize actionable insights.

---

🔍 Project Overview

Customer churn is one of the most critical problems in the telecom industry. This project builds a full-stack churn analysis system that:

1. Ingests & cleans raw customer data using SQL Server
2. Trains ML models (Logistic Regression, Random Forest, XGBoost) in Python to predict churn
3. Visualizes churn trends and predicted churners through an interactive Power BI dashboard

---

🏗️ Project Architecture

SQL Server  ──►  Python (ML)  ──►  Power BI Dashboard
   │                  │                   │
   │  Data cleaning   │  Model training   │  Churn summary
   │  & staging       │  & prediction     │  & forecasting

---

📁 Repository Structure

churn-prediction-system/
│
├── data/
│   └── Customer_Data.csv              # Raw customer dataset
│
├── notebooks/
│   └── churn_prediction_model.ipynb   # ML model training (LR, RF, XGBoost)
│
├── sql/
│   └── churn_queries.sql              # Data exploration, cleaning & views
│
├── powerbi/
│   └── ChurnAnalysis.pbix             # Power BI dashboard file
│
├── docs/
│   └── Power_Query_and_DAX.md         # Power Query steps & DAX measures
│
├── assets/
│   └── images/                        # Dashboard screenshots & icons
│       ├── Summary.PNG
│       └── Prediction.PNG
│
├── requirements.txt
├── .gitignore
└── README.md

---

🧰 Tech Stack

Layer	Tools
Data Storage	SQL Server
Data Processing	SQL (T-SQL)
Machine Learning	Python, scikit-learn, XGBoost
Visualization	Power BI
Environment	Jupyter Notebook / Anaconda

---

⚙️ Setup & Usage

1. SQL – Data Preparation

Run the queries in sql/churn_queries.sql in the following order:

1. Data Exploration — check distinct values and null counts in stg_Churn
2. Data Cleaning — replace nulls and load clean data into prod_Churn
3. Create Views — create vw_ChurnData and vw_JoinData for Power BI

2. Python – ML Model Training

Install dependencies:

pip install -r requirements.txt

Run the notebook:

Open notebooks/churn_prediction_model.ipynb in Jupyter Notebook and run all cells.

The notebook covers:

• Data loading and label encoding
• Train/test split (80/20)
• Training and evaluating three models:
  • Logistic Regression
  • Random Forest (200 estimators)
  • XGBoost (300 estimators, learning rate 0.05)
• Exporting predictions for Power BI

3. Power BI – Dashboard

Open powerbi/ChurnAnalysis.pbix in Power BI Desktop.

Power Query transformations and all DAX measures are documented in docs/Power_Query_and_DAX.md.

---

📊 Dashboard Preview

Summary Page

Churn Prediction Page

---

📈 Model Results

Three classification models were trained and compared on the customer dataset:

Model	Notes
Logistic Regression	Baseline model
Random Forest	200 estimators, best interpretability
XGBoost	300 estimators, highest accuracy

---

📂 Dataset

The dataset (data/Customer_Data.csv) contains customer-level telecom data including demographics, services subscribed, billing information, and churn status.

Key columns: Customer_ID, Gender, Age, Tenure_in_Months, Internet_Service, Contract, Monthly_Charge, Total_Revenue, Customer_Status

---

🤝 Contributing

Pull requests are welcome. For major changes, please open an issue first to discuss what you would like to change.

---

📄 License

This project is for educational and portfolio purposes.