Explainable Time Series Prediction of Tyre Energy in Formula One Race Strategy

Deep learning models successfully predicted tyre energy during F1 races using real telemetry data from Mercedes-AMG PETRONAS, demonstrating that machine learning can forecast the physical degradation of racing tyres with "impressive performance"—a key input for optimizing pit stop timing and tyre compound selection.

Explainability methods revealed the reasoning behind predictions, with the researchers using feature importance and counterfactual explanations to understand which factors (such as driving behavior or track conditions) most influenced the model's forecasts, making the AI system's decisions transparent rather than a "black box."

Decision tree-based models (XGBoost) matched deep learning performance, suggesting that simpler, more interpretable machine learning approaches can be competitive with complex neural networks for this application, which is important for real-world adoption where teams need to understand and trust their strategic recommendations.

The work demonstrates practical responsible AI in high-stakes decision-making, showing how explainability techniques can be integrated into automated systems that assist human experts (race strategists) rather than replacing them, allowing teams to verify recommendations before making split-second decisions that affect race outcomes.

Automated strategy assistance could reduce human error in time-critical environments, where the paper suggests their method could help F1 teams optimize decisions during races—a domain where even small improvements in tyre management directly translate to competitive advantage.