📉 Customer Churn Prediction System

An end-to-end, production-ready Machine Learning system designed to predict customer churn in subscription-based businesses such as Telecom, SaaS, and Banking.

🚀 Access Live Web Application

---

🖥️ Application Preview

Live Interactive Web Dashboard deployed on Streamlit Cloud

---

🔗 Live Application Link

🌐 Live Web App: https://churnmodel-by-vishal-dubey.streamlit.app/

Click the link above to access the live app, adjust customer profile parameters, predict churn risk probabilities in real-time, and view tailored business retention recommendations.

---

🚀 Project Overview

Customer churn directly impacts business revenue. Failing to identify churn-prone customers leads to significant financial loss, while timely detection enables proactive retention strategies.

🎯 Objective

Predict whether a customer will churn (1) or stay (0), enabling businesses to take data-driven retention actions before losing valuable accounts.

---

🧠 Key Highlights

• ✅ End-to-End ML Pipeline: Complete flow from raw data EDA to interactive web deployment.
• ✅ Data Leakage Prevention: Removed post-churn features like total charges and location noise to ensure high real-world accuracy.
• ✅ Class Imbalance Management: Handled ~26% baseline churn distribution effectively.
• ✅ Recall-Optimized: Prioritized high recall for churners to minimize costly false negatives.
• ✅ Ensemble Learning: Utilized a Soft Voting Classifier combining Logistic Regression, Random Forest, and Gradient Boosting.
• ✅ Interactive Dashboard: Built with Streamlit offering real-time prediction sliders, risk badges, and business suggestions.

---

🧩 Solution Strategy

The project follows a structured, industry-aligned ML workflow:

1. Exploratory Data Analysis (EDA): Deep dive into customer demographics, contract types, and billing behaviors.
2. Hypothesis Testing: Validated core assumptions regarding contract duration, monthly charges, and churn rate.
3. Data Leakage Prevention: Filtered out collinear features (TotalCharges, TotalRevenue) available only after customer lifecycle ends.
4. Feature Engineering & Encoding: Applied categorical target encoding and standard scaling.
5. Baseline Modeling: Trained Logistic Regression as a baseline model.
6. Advanced & Ensemble Modeling: Built Random Forest, Gradient Boosting, and a Soft Voting Classifier.
7. Business Metrics Optimization: Tuned decision thresholds focusing on recall over raw accuracy.
8. Deployment: Packaged and deployed via Streamlit Cloud for instant accessibility.

---

📂 Dataset Description

• Dataset Size: ~7,000+ customer records
• Domain: Telecom customer behavior

🎯 Target Variable

• ChurnLabel
  • 1 → Customer Churned ❌
  • 0 → Customer Retained ✅

---

🧾 Feature Categories

Category	Key Features
👤 Demographics	Age, Gender, Senior Citizen Status, Dependents
📊 Account & Contract	Contract Type (Month-to-month, One year, Two year), Tenure (Months)
💰 Usage & Billing	Monthly Charges, Payment Method, Internet Service Type
⭐ Customer Value	CLTV (Customer Lifetime Value), Satisfaction Score

---

📊 Key EDA Insights

• 🔹 Contract Sensitivity: Customers on Month-to-month contracts exhibit the highest churn probability.
• 🔹 Tenure Impact: Low-tenure customers (< 12 months) are significantly more prone to churn.
• 🔹 Pricing Pressure: Higher monthly charges positively correlate with elevated churn risk.
• 🔹 Satisfaction Score: Low customer satisfaction scores serve as the strongest early warning indicator.

---

🤖 Models & Performance

Model	Accuracy	Recall (Churn)	ROC-AUC	Status
Logistic Regression	~88%	Good	~0.96	Baseline
Random Forest	~88%	Moderate	~0.95	Tree-based
Gradient Boosting	~95%	High	~0.99	Strong Performer
Voting Classifier (Ensemble)	~95%	Highest 🔥	~0.98	Production Model

---

🧠 Business Value & Metric Selection

In customer churn prediction, a false negative (missing a churner) is far more expensive than a false positive (sending a discount to a loyal customer). Therefore, our optimization strategy explicitly maximizes Recall for Churned Customers, ensuring retention teams can proactively intervene.

---

📁 Project Structure

Churn_Model/
│
├── assets/
│   └── app_preview.png         # Web Application Screenshot Showcase
├── data/                       # Dataset files
├── notebooks/                  # EDA & Model Training Jupyter Notebooks
├── model/
│   ├── churn_model.pkl         # Trained Soft Voting Ensemble Model
│   └── feature_names.pkl       # Feature Metadata
├── app.py                      # Main Streamlit Web Application
├── requirements.txt            # Python Dependencies
└── README.md                   # Documentation

---

⚡ How to Run Locally

1️⃣ Clone the Repository

git clone https://github.com/Vishaldubey2210/Churn_Model.git
cd Churn_Model

2️⃣ Install Dependencies

pip install -r requirements.txt

3️⃣ Run Streamlit Web Application

streamlit run app.py

The application will launch locally at http://localhost:8501.

---

🚀 Future Roadmap

• Integrate SHAP (SHapley Additive exPlanations) for local model interpretability.
• Automated Hyperparameter Optimization using Optuna.
• REST API development via FastAPI.
• CI/CD pipeline integration with GitHub Actions.

---

👨‍💻 Author

Vishal Kumar
Aspiring Machine Learning Engineer

• 🌐 Live Application: https://churnmodel-by-vishal-dubey.streamlit.app/
• 💻 GitHub: @Vishaldubey2210