A data science project to predict customer churn using behavioral metrics from a telecom provider. The project covers end-to-end steps from data exploration and feature engineering to model training and interpretation — with actionable business insights.
Customer churn is a key concern in the telecom industry due to its direct impact on revenue and customer lifetime value. In this project, we developed a machine learning model that classifies customers as likely to churn or not, based on usage patterns, demographics, and service interactions.
- Analyze telecom customer behavior through exploratory data analysis (EDA)
- Identify key drivers of churn
- Build and evaluate classification models
- Provide business recommendations based on model outputs
- Prepare the model for deployment and integration into CRM systems
The dataset includes:
- Customer demographics
- Call duration and frequency
- SMS usage
- Account status and complaints
- Subscription length and customer value
📁 Source: Provided as customer_churn.csv (confidential)
- Younger age groups favor SMS over calls
- Customers with complaints, short subscription history, and low usage are at higher churn risk
- High-value customers with stable, long-term usage patterns are more loyal
📊 Key Features Identified:
Status_1,Status_2(account flags)Seconds of Use,Frequency of UseCustomer Value,Subscription LengthComplaint,Distinct Called Numbers
| Metric (Churn Class) | Logistic Regression | Random Forest |
|---|---|---|
| Precision | 51.0% | 88.2% |
| Recall | 89.9% | 81.1% |
| F1-score | 65.0% | 84.5% |
| Accuracy | 84.9% | 95.3% |
✅ Random Forest chosen as final model for deployment due to its superior balance of precision and recall.
- True Positives (120): Correctly predicted churners
- False Positives (16): Non-churners wrongly flagged
- False Negatives (28): Missed churners
- True Negatives (781): Correctly predicted non-churners
Key predictors: service status, call duration, engagement level, and customer lifecycle.
- Save model using
joblib - Deploy with a simple Streamlit or Flask API
- Weekly scoring pipeline + CRM integration for automated retention targeting
- Focus on customers with complaints, low recent usage, or short tenure
- Launch early-intervention campaigns for new users
- Use prediction scores to prioritize support and loyalty rewards
- Python (Pandas, NumPy, Scikit-learn, Matplotlib, Seaborn)
- Jupyter Notebook
- Machine Learning: Logistic Regression, Random Forest
- Streamlit (for future deployment)
This project demonstrates how machine learning can turn raw behavioral data into real business impact. By predicting churn with high accuracy, companies can intervene earlier and reduce customer loss.
🔧 Ready to scale, interpret, and deploy in a business setting.