📊 Telecom Customer Churn Prediction

---

📌 Project Overview

Customer churn is one of the most critical problems in the telecom industry.
This project aims to predict customer churn using machine learning and provide actionable business insights for customer retention.

The workflow covers:

• Data exploration & visualization
• Data cleaning & preprocessing
• Model building (Decision Tree & XGBoost)
• Evaluation (Accuracy, F1, ROC AUC, Cross-Validation)
• Feature importance & insights
• Saving the best model for future use

---

📂 Dataset

• Source: IBM Sample Dataset (Telco Customer Churn)
• File used: Telco_customer_churn.xlsx
• Target variable: Churn (Yes → 1, No → 0)
• Size: ~7,000 customers, 20+ features

Features include:

• Demographics: Gender, SeniorCitizen, Partner, Dependents
• Services: InternetService, Streaming, OnlineSecurity, TechSupport
• Contracts: Contract type, Payment method, Paperless billing
• Financials: Tenure, MonthlyCharges, TotalCharges

---

⚙️ Methodology

1. Data Cleaning & Preprocessing

• Dropped identifiers and leakage columns (CustomerID, Churn Reason, etc.)
• Converted TotalCharges to numeric
• Handled missing values with imputation (median for numeric, most_frequent for categorical)
• Scaled numeric features
• One-hot encoded categorical features

2. Exploratory Data Analysis

• Visualized churn distribution (imbalanced dataset: ~26% churners)
• Explored key drivers: Tenure, Contract type, Monthly charges
• Correlation heatmap for numeric variables

3. Model Building

• Decision Tree (baseline): simple, interpretable
• XGBoost (advanced): robust gradient boosting model

Both models were wrapped inside sklearn Pipelines for clean preprocessing + modeling.

4. Evaluation

Metrics used:

• Accuracy
• Precision, Recall, F1-score
• ROC AUC
• 5-fold Cross-Validation (for XGBoost)

5. Feature Importance

• Extracted top churn drivers from XGBoost
• Visualized the top 15 most important features

---

🏆 Results

Model	Accuracy	F1 Score	ROC AUC
Decision Tree	~0.73	~0.62	~0.74
XGBoost	~0.82	~0.71	~0.85

• XGBoost outperformed the Decision Tree across all metrics.
• Top churn drivers (example): Contract type, Tenure, MonthlyCharges, PaymentMethod, InternetService.

---

📊 Visualizations

Here are some key visuals from the analysis:

• Churn Distribution
  

• Feature Importance
  

• ROC Curve
  

---

💡 Business Insights

1. Contract Type matters: Month-to-Month customers churn the most.
   → Offer discounts or loyalty benefits for longer-term contracts.
2. Early Tenure churn risk: Customers in their first year are more likely to leave.
   → Focus on onboarding experience and engagement campaigns.
3. High Monthly Charges: Customers with higher bills show higher churn.
   → Consider competitive pricing or bundling offers.

---

🚀 Tech Stack

• Python 3.9+
• Libraries: pandas, numpy, matplotlib, seaborn,
  scikit-learn, xgboost, joblib

---

📦 How to Run

# 1. Clone this repo
git clone https://github.com/VedikaSankhe/Telecom-Customer-Churn.git
cd Telecom-Customer-Churn

# 2. Install dependencies
pip install -r requirements.txt

# 3. Run the notebook
jupyter notebook notebooks/customerchurn.ipynb

👩‍💻 Author

Developed with ❤️ by Vedika Sankhe

If you like this project, don’t forget to ⭐ the repo!

---

💬 Feedback

If you have any feedback, please reach out at vedikasankhe11@gmail.com

---

📜 License

This project is licensed under the MIT License – see the LICENSE file for details.