📞 Telecom Customer Churn Prediction Engine

This project demonstrates a full-stack deployment of a Machine Learning model designed to predict customer churn risk in real time, focusing on optimizing high-value customer retention (Recall).

---

🎯 Project Goal

The primary objective was to minimize False Negatives (missed churners) to maximize revenue protection.
This was achieved by optimizing for high Recall on the minority (churn) class.

---

🧠 Final Model Performance

Metric	Description	Value
Model Used	Decision Tree Classifier
Key Optimization	class_weight='balanced'
Churn Recall (Capture Rate)	% of actual churners correctly identified	79%
Churn Precision (Efficiency)	% of predicted churners who actually churn	49%
Missed Customers (FN)	False Negatives — churners missed by model	80

This section highlights the core business trade-off:
maximizing Recall to minimize revenue loss, even if Precision decreases slightly.

---

⚙️ Technical Pipeline (MLOps)

The application transfers data between the frontend, backend, and model in three key stages:

1️⃣ Data Science & Feature Engineering

• Feature Creation: Built a robust 18-feature set, including metrics such as Charge Deviation and Number of Services.
• Imbalance Handling: Applied class_weight='balanced' during training to increase model sensitivity.
  • Result: Achieved a 79% Recall on the minority churn class.

---

2️⃣ Deployment Stack (API Backend)

• Model Artifact: Optimized Decision Tree saved as

  churn_model.pkl