🏎️ F1 Race Pace Predictor

The F1 Race Pace Predictor is a production-grade Machine Learning ecosystem designed to forecast Formula 1 Grand Prix outcomes. By integrating historical racing data from FastF1, real-time environmental conditions via OpenWeather API, and advanced ensemble models, the system provides high-fidelity predictions for driver lap times across various international circuits.

---

🌍 Live Deployment

The project is fully deployed and accessible via a custom domain with modern networking support:

• Live Link/Domain: https://f1predictor.tech
• Subdomain Support: Fully configured for both the root domain (@) and www subdomain.
• Networking: Dual-stack support for both IPv4 and IPv6 traffic.
• Infrastructure: Hosted on Render with automated CI/CD pipelines and SSL encryption.

---

🧠 Data Science & Feature Engineering

The core of this project lies in transforming raw telemetry and historical logs into high-impact features for predictive modeling.

The Data Pipeline

• Historical Extraction: Utilizes the FastF1 library to pull session-specific lap times and sector breakdowns from the 2024 season.
• Real-Time Weather: Ingests live API data from OpenWeather to factor in track temperature and rain probability.
• Feature Imputation: Employs SimpleImputer with a median strategy to handle missing historical sector data, ensuring model stability during real-time inference.

Advanced Feature Engineering

• Team Performance Score: A normalized metric derived by dividing current constructor points by the maximum points in the field, quantifying the "mechanical advantage" of each car.
• Clean Air Race Pace: A weighted feature that adjusts expected lap times based on a driver's ability to maintain pace without aerodynamic interference.
• Total Sector Time: A composite feature summing the three distinct track sectors to provide a granular view of car-track efficiency.

Machine Learning Models

The project utilizes a diversified ensemble approach to minimize error:

• Gradient Boosting Regressor (GBR): Used for the Mexico and US GP models to capture complex, non-linear relationships between qualifying speed and race endurance.
• XGBoost (XGBRegressor): Leverages monotone constraints to ensure that as qualifying speed increases, predicted race pace remains logically consistent.
• Deep Learning: A Feed-Forward Neural Network (FFN) built with TensorFlow for high-dimensional pattern recognition on the Abu Dhabi circuit.

---

🚀 Key Features

• Dynamic Inference Engine: A FastAPI backend that automatically loads specialized .joblib model artifacts on startup.
• Explainable AI (XAI): Integrated SHAP visualizations to provide transparency into how each feature affects the final prediction.
• Performance Benchmark: Optimized for high-availability, delivering sub-200ms inference response times even under concurrent load.

---

🔧 Setup & Installation

1. Clone the Repository

git clone [https://github.com/MallamTeja/f1_models.git](https://github.com/MallamTeja/f1_models.git)
cd f1_models

2. Local Execution

# Set up virtual environment
python -m venv venv
source venv/bin/activate  # Windows: venv\Scripts\activate

# Install dependencies
pip install -r req.txt
d

3. Running with SERVER

# Run the container
py -m uvicorn main:app --host 127.0.0.1 --port 8000

🧪 Testing Run the automated test suite to verify model integrity and API logic:

pytest test_main.py

This validates edge cases, including driver code verification and logical lap time ordering—ensuring predicted race pace is realistically slower than qualifying speed.

Disclaimer: This project is unofficial and is not associated in any way with the Formula 1 companies. F1, FORMULA ONE, FORMULA 1, FIA FORMULA ONE WORLD CHAMPIONSHIP, GRAND PRIX and related marks are trademarks of Formula One Licensing B.V.