Data Science · September 2026
Customer Churn Prediction
Predicting telecom churn and explaining why each customer is at risk.
- ROC-AUC
- 0.849
- Recall
- 0.807
Overview
End-to-end ML system on 7,043 Telco customers. After benchmarking Logistic Regression, Random Forest, XGBoost and LightGBM, the best performers were combined into a stacking ensemble, with SHAP explaining individual predictions. Deployed as a four-view Streamlit dashboard: portfolio summary with revenue at risk and a retention ROI simulator, what-if single prediction, validated batch scoring, and per-customer explanations.
Tech stack
- Python
- scikit-learn
- XGBoost
- LightGBM
- SHAP
- Streamlit