Formula1 — High-Accuracy Track Time Prediction

Analytics-First Machine Learning Project

Objective

Predict segment times, sector times, and total lap time at Spa-Francorchamps for an F1-style hybrid car with maximum achievable accuracy using physics-informed synthetic data and interpretable machine learning.

---

Design Parameter Vector (Fixed)

Each vehicle setup is defined by exactly 6 parameters:

Parameter	Symbol	Description	Range
Total Mass	m	Vehicle mass in kg	700–850 kg
Aero Load Coefficient	C_L	Normalized downforce coefficient	0.8–1.5
Aero Drag Coefficient	C_D	Normalized drag coefficient	0.7–1.3
Electric Power Fraction	α_elec	Fraction of power from electric (0–1)	0.0–0.4
Max Deployable Energy	E_deploy	Max electric energy per lap (MJ)	2.0–4.0 MJ
Cooling Aggressiveness	γ_cool	Cooling factor affecting power availability	0.8–1.2

---

Output Targets

The model predicts:

y = [
    segment_times[1..N],   # N ≈ 20 Spa segments
    sector_times[1..3],    # 3 sectors
    lap_time               # Total lap time
]

---

Project Structure

Formula1/
├── src/
│   ├── data/           # Synthetic data generation
│   ├── features/       # Feature engineering
│   ├── models/         # Model training & tuning
│   ├── evaluation/     # Evaluation & metrics
│   └── analysis/       # Sensitivity & robustness analysis
├── data/
│   ├── raw/            # Raw track/physics data
│   ├── processed/      # Preprocessed datasets
│   └── synthetic/      # Generated training data
├── models/
│   ├── checkpoints/    # Training checkpoints
│   └── final/          # Production-ready models
├── outputs/
│   ├── figures/        # Plots and visualizations
│   └── reports/        # Evaluation reports
├── notebooks/          # Exploratory analysis
├── configs/            # Configuration files
└── requirements.txt    # Dependencies

---

Execution Order

1. Finalize physics-informed data generator
2. Generate large, diverse datasets (≥100,000 samples)
3. Engineer derived features (power-to-weight, aero efficiency, etc.)
4. Train baseline models (Linear, Ridge, Lasso)
5. Train advanced models (XGBoost, LightGBM, Random Forest)
6. Deep evaluation & sensitivity analysis
7. Freeze best-performing, explainable model

---

Key Principles

• Accuracy over speed: Prioritize prediction quality
• Physics-informed: All assumptions grounded in engineering logic
• Explainable: No black-box acceptance without reasoning
• Robust: Stable across retraining and edge cases

---

Quick Start

# Create virtual environment
python -m venv venv
source venv/bin/activate  # On Windows: venv\Scripts\activate

# Install dependencies
pip install -r requirements.txt

# Generate synthetic data
python -m src.data.generator

# Train models
python -m src.models.train

# Evaluate
python -m src.evaluation.evaluate

---

License

MIT License