F1NN-MonteCarlo: An F1 Racing Strategy Prediction System Based on Neural Networks and Monte Carlo Simulation

F1NN-MonteCarlo is an end-to-end F1 racing strategy prediction framework that integrates the FastF1 data engineering pipeline, neural network lap time prediction, and Monte Carlo simulation. It aims to provide data-driven strategy recommendations for team decisions. This project supports F1 racing strategy formulation through quantitative analysis, changing the traditional pattern of relying on empirical judgment.

The complexity of Formula 1 racing strategy lies in making optimal decisions (such as pit stop timing and tire compound selection) under rapidly changing track conditions, which directly affect race results. Traditional strategies often rely on empirical judgment, while modern F1 data analysis is changing this pattern. MahirK7's graduation project, F1NN-MonteCarlo, addresses this need by building a complete data engineering and machine learning framework that combines neural network prediction and Monte Carlo simulation to provide quantitative analysis support.

This project adopts a modular end-to-end architecture covering the complete process from data collection to strategy generation:

Data Layer: Uses the FastF1 library to extract official F1 race data (lap times, pit stop records, tire usage, weather, etc.), which is cleaned and stored in a PostgreSQL database. Feature Engineering Layer: Extracts key features such as driver historical performance, track characteristics, and tire degradation curves. Prediction Layer: Implements three neural network architectures—MLP, LSTM, and TCN—to predict lap times, and selects the optimal model through comparison. Simulation Layer: Embeds the trained model into a Monte Carlo simulation framework to evaluate the expected performance of different strategies. Interaction Layer: Builds a visual dashboard based on Streamlit, supporting users to select races and run simulations.

The project's tech stack follows modern data engineering best practices: Uses Apache Airflow to orchestrate processes (supporting manual triggering and scheduled execution), and Docker Compose to deploy PostgreSQL and Airflow to ensure environment consistency. Model training focuses on accuracy, interpretability, and stability; training results, weights, and visual charts are archived in the output directory for easy tracking. Monte Carlo simulation is the core innovation—it simulates uncertain factors such as safety cars, weather changes, and driver errors through random sampling to evaluate strategy robustness, which is more aligned with the complexity of real races than deterministic models.

A highlight of the project is comparing the generated strategies with actual 2025 race data for validation (backtesting), which not only tests the model's predictive ability but also verifies the practical effectiveness of the strategy recommendations, allowing quantification of the system's strategy optimization potential.

For data science learners, this project demonstrates how to integrate multiple technical components into a complete solution (from data extraction, cleaning, and storage to feature engineering, model training, and simulation optimization) with clear module division. The code structure follows Python engineering standards, using requirements.txt to manage dependencies and .env to configure environment variables, reflecting production-level development standards.

The framework has good scalability: In the future, real-time data streams can be integrated to enable dynamic strategy adjustments during races; reinforcement learning can be introduced to allow the system to continuously learn and optimize from strategy execution results. It is a reference project worth in-depth study for F1 enthusiasts and data science practitioners.