End-to-End Customer Churn Analysis & Prediction

An end-to-end analytics project that identifies key drivers of customer churn and predicts future churn using a multi-tool stack. This repository contains the SQL scripts for ETL, a Python notebook for machine learning, and a link to the final interactive Power BI dashboard.

---

📊 Live Power BI Dashboard

The final output is an interactive dashboard that provides a high-level summary of churn metrics and a detailed view of customers predicted to churn.

>> Click Here to View the Live Dashboard <<

---

🎯 Project Goal

The primary objective of this project was to analyze a telecom's customer dataset to understand the primary factors leading to churn. This was followed by building a predictive model to identify at-risk customers, enabling the business to take proactive retention measures.

---

⚙️ Architecture & Tech Stack

This project follows a realistic, multi-stage analytics workflow:

1. Database & ETL (SQL Server):

   • Raw data was loaded into a SQL Server database.
   • SQL scripts were used for the initial ETL process: cleaning nulls, standardizing categorical data, and performing transformations.

2. Predictive Modeling (Python):

   • A Jupyter Notebook was used for exploratory data analysis (EDA) and to build the machine learning model.
   • Libraries: Pandas, NumPy, Scikit-learn, Matplotlib, Seaborn.
   • A Random Forest Classifier was trained to predict churn probability for each customer.

3. Visualization & Reporting (Power BI):

   • Connected to the cleaned SQL database to build the main summary report.
   • Imported the prediction results from the Python model to create an "Actionable Insights" page, listing high-risk customers.
   • Key Features: DAX Measures, Interactive Slicers, Bookmarks, and Drill-throughs.

---

📈 Project Workflow & Key Steps

1. ETL & Data Cleaning (SQL)

The raw data was first loaded into a SQL Server table. The following cleaning and transformation steps were performed using SQL:

-- Example: Creating Views for Power BI
Create View vw_ChurnData as
	select * from prod_Churn where Customer_Status In ('Churned', 'Stayed')


Create View vw_JoinData as
	select * from prod_Churn where Customer_Status = 'Joined'

(The full script can be found in the /sql/ directory.)

2. Machine Learning Model (Python)

After cleaning the data, a predictive model was built to identify customers likely to churn.

• Feature Engineering: Created new features and encoded categorical variables for the model.
• Model Selection: A Random Forest Classifier was chosen for its high accuracy and ability to handle complex interactions between features.
• Evaluation: The model achieved the following performance on the test set:
  • Accuracy: 71%
  • Precision: 78%
  • Recall: 65%

# Example: Training the Random Forest model
from sklearn.ensemble import RandomForestClassifier

# Instantiate model with 100 decision trees
rf_model = RandomForestClassifier(n_estimators=100, random_state=42)

# Train the model on training data
rf_model.fit(X_train, y_train)

(The full notebook can be found in the /notebook/ directory.)

---

🚀 How to Run

To replicate this project, you will need to:

1. Set up a SQL Server instance and import the raw data from the /data/raw folder.
2. Run the script in the /sql/ folder to explore/clean the data and to create views as well.
3. Execute the Jupyter Notebook in /notebook/ to generate the prediction CSV.
4. Open the .pbix file and refresh the data sources to point to your local SQL database and the generated prediction file.

---

💡 Future Improvements

• Deploy the machine learning model as a REST API for real-time predictions.
• Automate the entire pipeline using a workflow orchestrator like Apache Airflow.
• Migrate the database and pipeline to a cloud platform like GCP, Azure or AWS.

---