📉 Telecom Customer Churn — End-to-End ML Pipeline
Production-style churn prediction workflow with SHAP explainability and profit-driven evaluation.
Project Overview
Built on the IBM Telco dataset, this project delivers an end-to-end churn prediction system: from data acquisition & cleaning through feature engineering, model selection & evaluation, to a Streamlit app deployment. It incorporates leakage prevention, profit-optimized decision thresholds, and SHAP-based interpretability for actionable insights.
Key Features
- Automated data preprocessing — normalization, duplicate removal, imputation, and leakage control.
- Feature engineering — e.g.,
contract_length_months,is_electronic_check, and tenure-based bins. - Model comparison — Logistic Regression, Random Forest, XGBoost (selected for deployment).
- Profit-optimized threshold — uses outreach costs, incentives, and margins in evaluation.
- SHAP explainability — global and per-customer feature impact visualizations.
- Streamlit app — batch scoring, single-customer probability calculator, and SHAP insights.
Tech Stack ⚙️
- Python ·
pandas,scikit-learn,xgboost,shap - Data ingestion via
kagglehub - Visualization:
matplotlib,Altair,Plotly
- Deployment: Streamlit
- Model persistence:
joblib - Environment:
venv(dependency pins for reproducibility)
Try It Yourself
python -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt
python -m src.models.train --csv "data/raw/Telco-Customer-Churn.csv"
streamlit run app/Home.py