Data-driven pit stop decision support for Formula 1 using deep learning models.

Area of Science:

• Motorsport Analytics
• Machine Learning in Sports
• Data-Driven Decision Making

Background:

• Formula 1 racing success hinges on strategic decisions, particularly pit stop timing.
• Human judgment for pit stops is unreliable under dynamic race conditions.
• Leveraging raw telemetry data offers potential for objective, data-driven strategies.

Purpose of the Study:

• To develop and evaluate a data-driven framework for predicting optimal Formula 1 pit stop timings.
• To compare the performance of various deep learning architectures for this prediction task.
• To enhance the robustness and accuracy of pit stop timing predictions using advanced data techniques.

Main Methods:

• Utilized raw telemetry data from the FastF1 API.
• Implemented data preprocessing techniques: normalization, imputation, and Synthetic Minority Over-sampling Technique (SMOTE) for class balancing.
• Trained and evaluated five deep learning models: Bi-LSTM, TCN-GRU, GRU, InceptionTime, and CNN-BiLSTM.
• Assessed model performance using precision, recall, and F1-score metrics.

Main Results:

• The Bi-LSTM model demonstrated superior performance compared to other architectures.
• Bi-LSTM achieved a precision of 0.77, recall of 0.86, and an F1-score of 0.81 on the test set.
• The model's effectiveness was attributed to its ability to capture long-range temporal dependencies.
• A visualization interface was developed to display model predictions.

Conclusions:

• Deep learning, specifically the Bi-LSTM model, provides a robust and accurate method for predicting optimal Formula 1 pit stop timings.
• The data-driven approach significantly outperforms traditional human judgment in competitive racing scenarios.
• This framework offers a valuable tool for race strategy optimization in motorsports.