Back to case studies
ML · Data Science

XGE

End-to-end XGBoost pipeline predicting vehicle CO₂ emissions with feature engineering, SHAP explainability, and an interactive dashboard.

XGE 1XGE 2

Problem

Regulators and OEMs need a transparent way to estimate emissions for a given vehicle spec without trusting a black box.

Approach

  1. Cleaned EPA + Government of Canada emissions data; engineered 18 features.
  2. Tuned XGBoost via Optuna; benchmarked against LightGBM and CatBoost.
  3. Wired SHAP into the dashboard so each prediction shows the top 5 contributing features.

Results

R²
0.94
MAE
8.7 g CO₂/km
Models compared
3