Telecom Customer Churn Analytics & Retention Pipeline
Machine learning churn risk engine achieving 93.51% retention prediction accuracy and 75.25% F1-score
Gradient Boosting Classifier (500 Estimators)
Retention Test Accuracy: 93.51%
F1-Score: 75.25% | 5-Fold CV: 94.99%
4,617 subscribers (656 churned = 14.2%)
19 behavioral & account attributes
Target: Churn (True / False)
93.51% Test Accuracy & 75.25% F1-Score on unseen customer cohorts
Verified Notebook Results
Technologies & Machine Learning Frameworks
01Business & Clinical Problem Context
02Data Pipeline & Model Engineering
03Algorithm Evaluation & Empirical Benchmarks
| Algorithm / Model | Test Accuracy / Score | 5-Fold CV | F1-Score / Metric | Status |
|---|---|---|---|---|
| Gradient Boosting (Tuned) | 93.51% | 94.99% | 75.25% | best |
| XGBoost Classifier | 92.29% | 92.99% | 73.90% | contender |
| AdaBoost Classifier | 92.03% | 95.11% | 68.71% | contender |
| Random Forest (2000 Trees) | 90.82% | 92.36% | 69.36% | contender |
| Bagging Classifier | 87.45% | 92.68% | 60.49% | baseline |
| Decision Tree | 83.90% | 86.23% | 56.54% | baseline |
| Logistic Regression | 76.28% | 75.82% | 48.30% | baseline |
With a 14.2% baseline churn rate, a naive model that predicts zero churn achieves 85.8% accuracy while failing 100% of business retention objectives. By training with SMOTE oversampling and tuning decision thresholds, Gradient Boosting achieved 93.51% overall accuracy and a high 75.25% F1-score. XGBoost also excelled with 92.29% accuracy and 73.90% F1-score, confirming powerful non-linear separation on churn signals.
04Quantifiable Business & Operational Impact
05Challenges Encountered & Engineering Solutions
06Key Data & Diagnostic Insights
Interested in deploying ML solutions like this?
Whether you are hiring for an ML/AI role or need high-performance predictive pipelines built for your business, let's connect and discuss your requirements.