F1 StratLab Strategy Models
The machine learning models behind F1 StratLab, an open-source multi-agent system for Formula 1 race strategy. Six LangGraph sub-agents and a ReAct orchestrator call these models to produce pit-stop recommendations, tire-degradation forecasts, overtake and undercut probabilities, and answers grounded in the FIA regulations.
Links:
- Project: https://f1stratlab.com
- Documentation: https://docs.f1stratlab.com
- Source code: https://github.com/VforVitorio/F1-StratLab
- Dataset: https://huggingface.co/datasets/VforVitorio/f1-strategy-dataset
Models
| Task | Algorithm | Metric |
|---|---|---|
| Lap-time prediction | XGBoost | MAE 0.392 s |
| Tire degradation | TCN with Monte Carlo Dropout | P10/P50/P90 quantiles, pit-window detection |
| Overtake probability | LightGBM | AUC-ROC 0.876 |
| Safety-car probability | LightGBM | classifier |
| Pit-stop duration | HistGradientBoosting (quantile) | MAE 0.487 s |
| Undercut success | LightGBM (binary) | AUC-ROC 0.771 |
| Team-radio NLP | Whisper, RoBERTa, SetFit, BERT-large | 4-stage pipeline |
Training data
Trained on telemetry, lap data and race-control messages from 71 Grand Prix across the 2023 to 2025 seasons, taken from the FastF1 and OpenF1 public APIs. The processed data is published as a companion dataset: VforVitorio/f1-strategy-dataset.
Intended use
Research and educational use for Formula 1 strategy analysis. Not affiliated with Formula 1, the FIA or any team. Predictions are estimates, not guarantees.
Citation
@misc{vegasobral2026f1stratlab,
author = {Vega Sobral, V{\'i}ctor},
title = {F1 StratLab: AI Models for Strategy Recommendations in Formula 1 Races},
year = {2026},
note = {Bachelor's Thesis, Intelligent Systems Engineering, UIE Campus Coru{\~n}a},
url = {https://f1stratlab.com}
}
License
Apache 2.0. Author: Víctor Vega Sobral (https://github.com/VforVitorio).