📞 telecom-customer-churn-prediction

A machine learning project to identify customers likely to churn, generate churn risk scores, and provide actionable business insights for proactive retention strategies.

---

🎯 Project Objectives

• Identify customers at risk of churn
• Generate churn probability scores
• Segment customers into risk groups
• Understand key drivers of churn
• Provide business recommendations for retention

---

📊 Dataset

The dataset is loaded from a telecom SQL database and contains customer usage patterns, plan details, service interactions, and churn labels.

Key Features

• Call minutes & charges (Day, Evening, Night, International)
• Customer service calls
• Voicemail & international plans
• Account length & usage behavior
• Target variable: CHURN_FLAG

---

⚙️ Tech Stack

• Python
• Pandas, NumPy
• Scikit-learn
• Seaborn, Matplotlib
• PyMySQL
• Jupyter Notebook

---

🔎 Project Workflow

✅ Data Collection

• Loaded telecom customer data from SQL database

✅ Data Cleaning

• Standardized churn labels
• Converted object columns to numeric
• Handled missing values with median imputation

✅ Feature Engineering

• Created binary churn flag
• One-hot encoded categorical variables
• Scaled numerical features

✅ Modeling

Two models were trained:

• Logistic Regression (baseline)
• Random Forest (final model)

✅ Evaluation

Metrics used:

• Accuracy
• Precision
• Recall
• F1 Score
• ROC-AUC

---

⭐ Model Results

Logistic Regression

• Accuracy: 86%
• Recall (churn): 23%
• ROC-AUC: 0.79

Random Forest (Final Model)

• Accuracy: 94%
• Recall (churn): 60%
• ROC-AUC: 0.89

👉 Random Forest selected as best model.

---

🔥 Key Churn Drivers

Top predictors identified:

1. Customer Service Calls
2. Day Minutes & Charges
3. International Plan
4. Evening Minutes & Charges
5. International Usage

---

🎯 Customer Risk Segmentation

Customers classified into:

• High Risk → churn score ≥ 0.70
• Medium Risk → churn score 0.40–0.69
• Low Risk → churn score < 0.40

---

💼 Business Recommendations

• Improve customer support for high call-frequency customers
• Optimize pricing for heavy usage customers
• Reevaluate international plan offerings
• Implement proactive retention campaigns using churn scores
• Use segmentation for targeted marketing

---

🚀 How to Run

# Clone repository
git clone https://github.com/SyedHussain23/telecom-customer-churn-prediction.git

# Navigate to project folder
cd telecom-customer-churn-prediction

# Install dependencies
pip install -r requirements.txt

# Run notebook
jupyter notebook telecom-customer-churn-prediction.ipynb

---

⚠️ Database Setup

Update SQL credentials before running:

HOST = "your_host"
USER = "your_user"
PASSWORD = "your_password"
DB_NAME = "your_database"

👉 For security, store credentials using .env file.

---

📂 Project Output

The project generates:

• Churn predictions
• Churn probability scores
• Risk segmentation
• Business-ready dataset

telecom_churn_predictions.csv

---

🔮 Future Improvements

• Hyperparameter tuning
• Advanced ensemble models (XGBoost, LightGBM)
• Deep learning churn modeling
• Real-time churn prediction pipeline
• Deployment via API or dashboard

---

👨‍💻 Author

Syed Hussain Abdul Hakeem

🔗 GitHub: https://github.com/SyedHussain23

🔗 LinkedIn: https://www.linkedin.com/in/syed-hussain-abdul-hakeem

---

⭐ Support

If you found this project useful, please consider giving it a ⭐ on GitHub.