📉 Customer Churn Prediction

A complete end-to-end machine learning pipeline to predict customer churn using a high-dimensional real-world dataset.
This project covers data preprocessing, EDA, model development, rigorous evaluation, and an interactive web app built with Streamlit.

🧠 Problem Statement

The goal is to accurately predict if a customer will churn, using anonymized behavioral data.

Dataset Overview:

• ~167,000 records
• 215 numerical features (X0 to X214)
• Target variable: Target_ChurnFlag (0 = Not Churned, 1 = Churned)

🔍 Exploratory Data Analysis (EDA)

• Class Distribution: 60% retained, 40% churned
• Missing Data: Handled via median imputation
• Noise Removal: Non-numeric anomalies like "14 month lease" removed
• Feature Correlation: Validated with heatmaps and KDE plots

⚙️ Modeling Workflow

• Preprocessing: Stratified 80:20 split, median imputation
• Validation: 5-Fold Cross-Validation
• Metrics Used: ROC-AUC (primary), Accuracy, F1 Score, Confusion Matrix

---

🧪 Models Compared

Model	Accuracy	F1 Score	ROC-AUC	Remarks
✅ Random Forest	100%	1.00	~1.00	Final model
XGBoost	100%	1.00	~0.999	Close second
Gradient Boosting	100%	1.00	~1.00	Equally strong
LightGBM	100%	1.00	~1.00	Slight warnings
Logistic Regression	~49%	0.49	~0.50	Weak baseline

🏆 Final Model: Random Forest Classifier

• Training Accuracy: 100%
• Test Accuracy: 100%
• ROC-AUC: ~1.0
• 5-Fold CV ROC-AUC: 0.9999999959
• Confusion Matrix: [[20020, 0], [1, 13383]]

🔬 Model Interpretability

• Top 5 Features: X16, X7, X4, X19, X123
• Used for UI simplicity — full model trained on all 215 features
• Techniques:
  • Feature Importance (Random Forest)
  • Correlation Heatmap
  • KDE plots

🚀 Streamlit Web App

You can run the app locally:

streamlit run churn_app.py

• Input: 5 most important features
• Output: Predicted churn probability (e.g., 13%)

🖥️ Streamlit App Screenshot

🧰 Tech Stack

• Python
• pandas, numpy, scikit-learn
• xgboost, lightgbm
• matplotlib, seaborn
• streamlit
• joblib (for model loading)

📁 Project Structure

Customer-Churn-Prediction/
├── churn_app.py
├── random_forest_model.pkl
├── Assignment1.ipynb
├── app_screenshot.png
├── requirements.txt
└── README.md

🎯 Key Learnings

• Detected and addressed model overfitting
• Balanced model performance with deployment simplicity
• Understood real-world ML model evaluation cycles
• Delivered insights through EDA and feature analysis
• Built a working UI with real-time prediction logic

🔮 Future Enhancements

• Add SHAP visualizations for interpretability
• Deploy app on Render or HuggingFace Spaces
• Map feature codes (X0–X214) to actual business features
• Add periodic retraining pipeline

🙋‍♂️ Author

Divyansh Gautam 📬 divyanshgautam0410@gmail.com