metadata
datasets:
- scikit-learn/churn-prediction
- aai510-group1/telco-customer-churn
language:
- en
- ar
Customer Churn Prediction
Predicting telecom customer churn using Random Forest & SMOTE to enable proactive retention strategies.
Problem
Predict which telecom customers are likely to churn to enable proactive retention strategies.
Results
| Model | Accuracy | ROC-AUC | Recall |
|---|---|---|---|
| Random Forest | 79% | 0.813 | 49% |
| Balanced RF | 78% | 0.816 | 45% |
| SMOTE + RF | 77% | 0.809 | 56% ✅ |
Key Insights
- High monthly charges is the top churn driver
- Month-to-month contracts have highest churn risk
- New customers (< 6 months) are most vulnerable
Business Impact
SMOTE model saves ~$14,000 more annually compared to baseline by identifying 28 additional at-risk customers.
Tools
Python | Scikit-learn | SMOTE | Pandas | Matplotlib