How youssefelhassini063-spec/powerco-churn-prediction works
This repository is a notebook-based churn-prediction workflow for PowerCo SMEs: exploratory analysis informs feature engineering, followed by model evaluation and business diagnostics. The README documents a cost-sensitive XGBoost approach, a 20% decision threshold, SHAP analysis, and an executive slide deck. Raw data is not included. Notebook internals were not sampled, so the graph follows documented stages and avoids assumptions about specific code-level calls.
Main entry points: 03_Model_Evaluation.ipynb Model evaluation · 01_EDA.ipynb Exploratory analysis · Predictive_Churn_Engine_for_PowerCo_SMEs.pdf Executive findings
Components 9
Data analysis
- Exploratory analysis
01_EDA.ipynb - Feature engineering
Churn modeling
- Model evaluation
03_Model_Evaluation.ipynb - Random Forest baseline
03_Model_Evaluation.ipynb - Penalized XGBoost
03_Model_Evaluation.ipynb - Threshold calibration
03_Model_Evaluation.ipynb
Business impact
- SHAP diagnostics
03_Model_Evaluation.ipynb - Executive findings
Predictive_Churn_Engine_for_PowerCo_SMEs.pdf
Other components
- BCG X datasets
Connections 8
- BCG X datasets → Exploratory analysis: provides data
- Exploratory analysis → Feature engineering: informs features
- Feature engineering → Model evaluation: provides features
- Model evaluation → Random Forest baseline: tests baseline
- Model evaluation → Penalized XGBoost: tests model
- Model evaluation → Threshold calibration: calibrates threshold
- Model evaluation → SHAP diagnostics: extracts values
- SHAP diagnostics → Executive findings: informs findings