Predicting Formula 1 Tire Degradation

Introduction and Motivation

Formula 1 is defined by milliseconds. Tire performance is a critical factor influencing lap times, pit stop strategy, and overall race outcomes. Yet, understanding when a tire will "fall off the cliff" is often part science, part art.

The goal of this project is to build a predictive model that can estimate how many laps into a stint a tire will begin degrading significantly. We do this by combining lap timing, tire metadata, driver+team info, and weather data across multiple seasons of F1 racing.

Data Collection and Labeling

Using the FastF1 Python library, we collected detailed lap-by-lap data for every race from the 2022 to 2025 F1 seasons (up to the 2025 Canadian GP). Each lap record includes lap time, compound, tire age, driver+team stats, track info, weather readings, and session metadata.

For each stint, we label the lap at which performance degradation begins. This is done by identifying when lap times exceed a dynamic threshold calculated as:

Only stints unaffected by retirements or collisions are included.

Feature Engineering

A wide variety of features are engineered from the raw data:

• Initial stint characteristics (compound, fresh tire status, starting lap, starting position)
• Lap time dynamics (average of first 3 laps, average of last 3 laps, delta between the two)
• Weather: air temperature, humidity, rainfall, wind speed and direction
• Context: number of nearby pit stops, prior pit stops, tire age at start
• Aggregated metrics: average tire life per driver and per team
• Driver/Team Info: driver and team information, including engine manufacturer

Modeling Approach

The target variable is tire_life — the number of laps into a stint before degradation. We applied a PowerTransformer to normalize its distribution (because tire degradation is not simply linear).

We trained and tuned 5 regressors using OptunaSearchCV for hyperparameter optimization:

• Random Forest Regressor
• Gradient Boosting Regressor
• XGBoost
• LightGBM
• CatBoost

Each model is wrapped in a Pipeline including preprocessing steps (e.g., one-hot encoding for categorical features).

Results

Performance metrics (MAE):

The best model (XGBoost) predicts tire degradation within 1.7 laps on average.

Conclusion

This project demonstrates that it is possible to model tire degradation in Formula 1 with fairly high accuracy using open telemetry data. Integrating this into simulators or strategy planners could benefit both race engineers and fans.

Code & Reproducibility

All code, data processing functions, and model training pipelines are available here: GitHub Repository.