F1 Qualifying Pace Predictor: ML Telemetry Pipeline
/ 1 min read
Table of Contents
Project Links
Project Overview
Developed a machine learning pipeline that predicts F1 Qualifying results based on Practice Session 3 (FP3) telemetry data. This project demonstrates that raw inputs like speed, throttle, and braking contain hidden signals correlating to qualifying performance, even under varying fuel loads.
Key Insights & Impact
Data Engineering (ETL): Leveraged the FastF1 API to extract granular telemetry at a 200Hz sample rate. Normalization Logic: Implemented a “Delta to Leader” calculation to normalize track evolution across different circuits. Feature Engineering: Engineered contextual features, including telemetry aggregation (Max Speed, Avg Throttle) and “Top Team” weighting. Predictive Modeling: Utilized a Gradient Boosting Regressor with a Learning-to-Rank approach to generate a performance score and predicted starting gri. Training Depth: Trained the model on historical race weekend data from the 2023-2024 seasons to ensure predictive accuracy.