Predicting the 2025 F1 Season with Machine Learning: From Data Collection to Race Outcome Forecasting

The 2025_f1_predictions project aims to use gradient boosting machine learning models and the FastF1 API, combined with historical race data and real-time qualifying information, to build an F1 race outcome prediction application. The project provides data-driven insights for racing enthusiasts, while offering practical cases and quantitative tools for data science learners and sports analysts.

For Racing Enthusiasts

• Enhance viewing experience: Understand drivers' winning probabilities before races
• In-depth race discussions: Analyze based on model outputs
• Verify prediction accuracy: Compare model results with actual races

For Data Science Learners

• End-to-end project practice: Cover the entire process from data collection to model deployment
• Time-series prediction practice: Handle data with time-series characteristics
• API integration experience: Obtain and process data from professional APIs

For Sports Analysts

• Quantitative analysis tools: Provide data support for subjective analysis
• Trend identification: Discover performance trends of drivers/teams
• Strategy evaluation: Analyze the impact of different strategies on outcomes

Data Collection Layer

Uses FastF1 API (Python library) to obtain the following data:

• Lap time data: Detailed lap time records for each driver
• Race results: Historical final rankings and results
• Telemetry data: Real-time vehicle performance metrics
• Qualifying information: Key data on grid positions for the main race

Data Processing Steps

1. Data cleaning: Handle missing values, outliers, and format issues
2. Feature engineering: Extract predictive features from raw data
3. Time-series alignment: Integrate time-series data from different sources
4. Normalization: Unify feature scales to ensure model fairness

Selected Gradient Boosting Machine (GBM) as the core algorithm, with advantages including:

• Handle complex non-linear relationships: Capture the correlation between driver performance and race outcomes
• Automatic feature selection: Iteratively identify important predictors
• High prediction accuracy: Outperforms single decision trees or linear models for structured data tasks
• Strong interpretability: Output feature importance rankings to understand influencing factors

Training Phase

The model learns patterns from historical data:

• Relationship between qualifying position and final ranking
• Impact of track characteristics on outcomes
• Historical performance trends of teams/drivers
• Correlation between weather conditions and race strategies

Prediction Phase

Executed after qualifying data is available:

1. Input the latest qualifying results
2. Convert to feature vectors understandable by the model
3. Output the probability distribution of drivers achieving specific positions
4. Generate final prediction results by synthesizing probabilities

Real-time Data Integration

The FastF1 API supports real-time updates, allowing the model to:

• Generate predictions immediately after qualifying ends
• Adjust parameters based on practice session performance
• Consider vehicle upgrades and track condition changes

Continuous Learning Mechanism

Continuously absorb new data during the season:

• Incremental training: Retain existing patterns and add new knowledge
• Performance monitoring: Track prediction accuracy and identify degradation
• Adaptive adjustment: Dynamically adjust prediction weights to adapt to changes in team performance

Current Limitations

• Difficulty predicting unexpected events: Random events like crashes or mechanical failures cannot be foreseen
• Weather dependence: Rainy races are heavily influenced by weather and strategies
• Adaptation to new rules: Correlation of historical data decreases when F1 introduces new rules

Future Improvements

• Multimodal data fusion: Integrate image data to enhance predictions
• Deep learning exploration: Try neural networks to handle time-series dependencies
• Uncertainty quantification: Provide confidence intervals for prediction results

The 2025_f1_predictions project demonstrates the application value of machine learning in the field of sports prediction. Through professional APIs, mature algorithms, and a clear architecture, it provides a practical prediction tool for F1 enthusiasts, while offering a full-process practical case for data science learners, covering data acquisition, feature engineering, and model training and deployment.