📊 Customer Churn Prediction – End-to-End ML Project

This project builds a complete Machine Learning pipeline to predict customer churn using real telecom data.

🔍 Objective

Predict whether a customer is likely to leave the telecom service (churn) based on their usage and profile.

---

📁 Dataset

Source: Telco Customer Churn - Kaggle
Rows: 7043 customers
Target Column: Churn (Yes/No)

---

🧠 Model: XGBoost Classifier

• Tuned with GridSearchCV
• Evaluated with Accuracy, Confusion Matrix, and Classification Report
• Explained using SHAP and LIME

---

⚙️ Pipeline Steps

1. Data Loading using Pandas
2. Data Cleaning & Preprocessing
   • Handled missing values
   • Encoded categorical variables
   • Converted TotalCharges to numeric
   • Target column: Churn → binary
3. Train/Test Split
4. Model Training with XGBoost + hyperparameter tuning
5. Model Interpretation
   • Global + Local explanations using SHAP
   • Individual prediction explained using LIME
6. Saving Outputs
   • Model saved as .pkl
   • Predictions saved as .csv and .db

---

📈 Results

• Best Parameters:
  n_estimators=50, learning_rate=0.1, max_depth=4

• Accuracy: 79.3%

• Classification Report:

Class	Precision	Recall	F1-Score
No Churn	0.83	0.90	0.87
Churn	0.65	0.49	0.56

---

📊 SHAP Interpretations

*Explains which features most impacted predictions.*

📊 LIME Interpretations

*local explanation for one specific prediction made by the model*

---

📦 Files in Repo

File	Description
churn_model.pkl	Trained XGBoost model
churn_predictions.csv	Saved test predictions
churn_results.db	SQLite version of predictions
Telco-Customer-Churn.csv	Raw dataset
churn_notebook.ipynb	Full end-to-end ML notebook
README.md	This project overview

---

✅ Skills Demonstrated

• Data Cleaning & Preprocessing
• Feature Engineering
• XGBoost + Hyperparameter Tuning
• Model Interpretation (SHAP + LIME)
• SQLite Integration
• Git & GitHub project documentation

---

🚀 Author

Sushma Sandanshiv
🔗 LinkedIn
💻 GitHub

---