🧠 Customer Churn Prediction

This repository presents a complete machine learning pipeline for customer churn prediction using an XGBoost classifier. The project includes data preprocessing, exploratory data analysis, model development, evaluation, and deployment via a Flask API.

---

🎯 Objective

• Develop a predictive model to identify customers likely to churn, enabling businesses to implement proactive retention strategies.
• Build and deploy a robust, scalable, and production-ready API for real-time inference.

---

📁 Project Structure

• model/
  Contains saved model artifacts:

  • xgb_model.pkl — Trained XGBoost classifier
  • scaler.pkl — StandardScaler object for feature normalization
  • threshold.pkl — Optimized decision threshold for classification

• app.py
  Flask application serving the model via a REST API

• requirements.txt
  List of required Python packages

• Customer_Churn_ML_Models.ipynb
  Jupyter Notebook detailing exploratory analysis, preprocessing steps, model training, and evaluation

---

⚙️ Model Development

🔄 Preprocessing

• Removal of irrelevant features (CustomerId, Surname)
• Encoding categorical variables (Geography, Gender)
• Outlier detection and treatment
• Feature scaling using StandardScaler
• Class imbalance handled with SMOTE

📊 Exploratory Data Analysis

• Gender: Female customers show higher churn (25.1%) than males (16.5%)
• Credit Card: Churn rate nearly identical regardless of credit card ownership
• Activity: Inactive members have significantly higher churn (26.9%) compared to active ones (14.3%)
• Geography: Customers from Germany churn the most (32%), nearly double the rate of those from France and Spain (~16%)

🧪 Model Training & Evaluation

Seven classification algorithms were evaluated:

• Logistic Regression
• K-Nearest Neighbors
• Decision Tree
• Random Forest
• Support Vector Machine
• Gaussian Naive Bayes
• XGBoost — selected as the final model based on superior performance

Evaluation Metrics:

• Accuracy: 0.849
• ROC AUC: 0.857
• Cross-validated ROC AUC: 0.961 ± 0.003
• Additional metrics: Precision, Recall, F1-score, Confusion Matrix
• Optimal threshold tuning (F1-maximizing threshold: 0.38) to balance precision and recall

---

🚀 Deployment

The trained model is deployed using Flask, allowing real-time predictions through a RESTful API. The API accepts customer data in JSON format and returns churn probability and prediction based on the optimized threshold.

---

▶️ How to Run

1. Install dependencies

   pip install -r requirements.txt
   

2. Start the Flask API

    ```bash python app.py ``` 

3. Make predictions

   Send a POST request to http://localhost:5000/predict with customer data in JSON format.

Example using cURL:

 ```bash
 curl -X POST http://127.0.0.1:5000/predict \
  -H "Content-Type: application/json" \
  -d '{
        "CreditScore": 600,
        "Age": 40,
        "Tenure": 5,
        "Balance": 50000,
        "NumOfProducts": 2,
        "EstimatedSalary": 100000,
        "Geography": "France",
        "Gender": "Male",
        "HasCrCard": 1,
        "IsActiveMember": 1
      }'
  

Sample API Response:

 ```bash
  {
  "churn_probability": 0.08523,
  "churn_prediction": 0,
  "threshold_used": 0.37769
  }

---

📜 License

This project is licensed under the MIT License. You are free to use, modify, and distribute this project with proper attribution.