Home

◆ BMW Luxury Sales Analytics — Project Wiki

Welcome to the complete documentation for BMW Luxury Sales Analytics — a production-grade analytics, econometrics and decision-intelligence platform for the BMW luxury-car market.

This project proves two things

1 — It can build a model that works. The pipeline reaches a cross-validated R² ≈ 0.85 on signal-bearing data, with SHAP recovering the true drivers. 2 — It won't fake it when the data is empty. The source dataset is structurally pristine but statistically signal-free; on it the same pipeline honestly scores R² ≈ 0 / AUC ≈ 0.5, proven with a permutation test and a positive control. Business value then comes from a labelled Scenario Simulator.

Predictive competence and intellectual honesty. That is the deliverable.

🔗 Quick links

🌐 Live dashboard	https://maxime2476-bmw-sales-analytics.hf.space
💻 Source code	https://github.com/maxime2476/bmw-sales-analytics
📖 Docs site (API ref)	https://maxime2476.github.io/bmw-sales-analytics/
🐳 Container image	ghcr.io/maxime2476/bmw-sales-analytics:latest
📊 Dataset	BMW Sales Dataset (Kaggle, eshummalik)

🗺️ How to read this wiki

Start with 01-Project-Overview for the why, then 02-Architecture for the how. The analytical heart is 04-Data-Integrity-and-Signal-Audit and its counterpart 05-Predictive-Capability. Everything else builds on those two.

Sections

• Foundations — 01-Project-Overview · 02-Architecture · 03-Data-Dictionary
• The analytical core — 04-Data-Integrity-and-Signal-Audit · 05-Predictive-Capability
• Analysis layers — 06-External-Data-Augmentation · 07-Econometrics · 08-Machine-Learning-and-Deep-Learning · 10-SQL-Analytics · 11-Explainability-SHAP
• Decision support — 09-Scenario-Simulator
• Product & engineering — 12-The-Dashboard · 13-Testing-and-Quality · 14-CI-CD-and-Deployment · 15-Reproducibility-and-Commands
• Reference — 16-Architecture-Decision-Records · 17-Results-Summary · 18-Limitations-and-Roadmap · 19-FAQ

⚡ 60-second tour

git clone https://github.com/maxime2476/bmw-sales-analytics
cd bmw-sales-analytics
make install-dev      # dependencies
make audit            # statistical signal audit (permutation + positive control)
make capability       # validated predictive-capability demo (CV R² ~0.85)
make sql              # DuckDB business-insights report
make test             # full suite behind a coverage gate
make app              # launch the dashboard → http://localhost:8501

---

Author: Maxime GOURGUECHON · MIT License · See 16-Architecture-Decision-Records for every significant decision.