# F1 Lap Time Prediction and Feature Analysis

> End-to-end machine learning framework for F1 lap time prediction using real-time telemetry data, achieving 94.8% R² through advanced feature engineering with track curvature, elevation profiles, and driver performance metrics.

- Author: Siddhant Shah
- Published: 2024-12-01
- Canonical URL: https://siddhants.com/projects/f1-lap-time-prediction/

## Overview

This project develops an end-to-end machine learning framework for predicting Formula 1 lap times using real-time telemetry data. Through sophisticated feature engineering and statistical analysis, we achieve 94.8% R² prediction accuracy.

**Repository**: [github.com/SidShah2953/F1-Telemetry-Analysis](https://github.com/SidShah2953/F1-Telemetry-Analysis)  
**Affiliation**: Boston University, Department of Computer Science  
**Program**: MS in Applied Data Analytics  
**Date**: December 2024

## Key Results

### Model Performance
- **R² Score**: 94.8% (exceptional predictive accuracy)
- **Features**: 20+ engineered variables
- **Methodology**: Advanced feature engineering with time-series quantization
- **Validation**: ANOVA and statistical hypothesis testing

### Top 5 Most Influential Features

| Feature | Importance | Description |
|---------|-----------|-------------|
| **Track Length** | 56.40% | Total circuit distance |
| **Elevation Std Dev** | 14.58% | Vertical terrain variation |
| **Total Elevation Change** | 6.23% | Cumulative altitude gain/loss |
| **Curvature Std Dev** | 5.17% | Track corner complexity |
| **Number of Corners** | 4.16% | Total turning points |

**Top 5 Total Contribution**: 86.54% of predictive power

## Project Objectives

### Primary Goal
Build a robust predictive model that can forecast F1 lap times based on:
- Track characteristics (geometry, elevation, surface)
- Driver performance patterns
- Car telemetry data
- Environmental conditions

### Research Questions
1. Which track features most significantly impact lap times?
2. How do driver styles differ in performance characteristics?
3. Can we predict lap times for new circuits?
4. What is the optimal feature set for prediction accuracy?

## Data Sources

### Telemetry Data
- **Real-time metrics**: Speed, throttle, brake, steering angle
- **Frequency**: High-resolution (millisecond-level)
- **Coverage**: Multiple seasons, all circuits
- **Volume**: Millions of data points

### Track Characteristics
- **Geometric Data**: GPS coordinates, corner angles, straight lengths
- **Elevation Profiles**: Altitude changes, gradients
- **Surface Data**: Track temperature, weather conditions
- **Configuration**: Circuit layout, sector boundaries

### Driver Performance
- **Historical Lap Times**: Race and qualifying data
- **Sector Times**: Granular performance breakdown
- **Team Information**: Constructor, car specifications
- **Session Conditions**: Practice, qualifying, race

## Feature Engineering

### Track Geometry Features

**Curvature Analysis**:
- Average curvature
- Standard deviation of curvature
- Maximum curvature (tightest corner)
- Curvature distribution (quantiles)

**Corner Characteristics**:
- Number of corners (total count)
- Corner complexity score
- Slow/medium/fast corner distribution
- Corner entry/exit angles

### Elevation Features

**Vertical Metrics**:
- **Elevation Standard Deviation** (14.58% importance)
- **Total Elevation Change** (6.23% importance)
- Maximum gradient (steepest climb/descent)
- Elevation gain vs. loss

**Impact**: Elevation changes affect:
- Engine power delivery
- Aerodynamic efficiency
- Tire wear patterns
- Driver energy management

### Track Length & Layout

**Distance Metrics**:
- **Track Length** (56.40% importance - dominant factor)
- Straight length (longest vs. average)
- Sector length distribution
- Track type (street vs. permanent circuit)

### Time-Series Quantization

**Technique**: Discretizing continuous telemetry signals
- Speed binning (low, medium, high)
- Throttle application quantiles
- Brake pressure zones
- Steering angle categories

**Benefits**:
- Captures non-linear relationships
- Reduces noise in telemetry data
- Enables pattern recognition
- Improves model generalization

### Driver-Specific Features

**Performance Metrics**:
- Historical lap time averages
- Qualifying vs. race pace differential
- Tire degradation patterns
- Consistency scores (lap time variance)

**Driving Style Indicators**:
- Aggressive vs. smooth braking
- Corner entry speed preferences
- Throttle application patterns
- Energy management strategies

## Statistical Analysis

### ANOVA (Analysis of Variance)

**Purpose**: Determine which features significantly impact lap times

**Methodology**:
- F-statistic calculation for each feature
- p-value analysis (significance testing)
- Effect size quantification
- Multiple comparison corrections

**Key Findings**:
- Track length: Highest F-statistic
- Elevation features: Statistically significant
- Curvature metrics: Strong predictive power

### Hypothesis Testing

**Null Hypothesis (H₀)**: Feature has no effect on lap time  
**Alternative Hypothesis (H₁)**: Feature significantly affects lap time

**Results**:
- Rejected H₀ for top 15 features (p < 0.01)
- Strong evidence for track geometry impact
- Validated feature selection methodology

### Multi-Dimensional Feature Interactions

**Complex Interactions Engineered**:
- Track length × curvature (handling vs. straight-line speed)
- Elevation change × number of corners (energy management)
- Driver consistency × track complexity
- Temperature × tire compound × track abrasiveness

These interactions capture non-linear effects that simple features miss.

## Driver Comparative Analysis

### Max Verstappen vs. Lando Norris

**Statistical Comparison** using multiple linear regression models to analyze driving styles and performance characteristics:

#### Track Complexity Management
**Verstappen Advantages**:
- More consistent performance on complex tracks
- Lower sensitivity to number of corners
- Lower sensitivity to maximum curvature
- More efficient adaptation to track characteristics

**Interpretation**: Superior technical skill in handling varied circuit types

#### Tire Management
**Verstappen Strengths**:
- Lower lap time degradation with tire wear
- More consistent performance across tire compounds
- Better tire preservation in race conditions

**Impact**: Strategic advantage in race simulations and pit stop strategy

#### Performance Factors
**Key Differences**:
- Different sensitivity to track temperatures
- Varying responses to weather conditions
- Distinct optimal setup preferences

**Application**: Team strategy optimization and car setup directions

## Machine Learning Model

### Algorithm Selection

**Model Type**: Gradient Boosting Regressor (XGBoost/LightGBM)

**Rationale**:
- Handles non-linear feature interactions
- Feature importance extraction
- Robust to outliers
- High predictive accuracy for tabular data

### Model Architecture

**Input Layer**: 20+ engineered features
**Training Strategy**: K-fold cross-validation
**Optimization**: Hyperparameter tuning (Grid Search)
**Validation**: Hold-out test set + temporal split

### Feature Importance Analysis

**Method**: SHAP (SHapley Additive exPlanations) values

**Insights**:
- Track length dominates (56.40%)
- Elevation features collectively contribute 20.81%
- Curvature metrics add 9.33%
- Driver-specific features: 8-10%

## Technical Implementation

### Python Stack

**Core Libraries**:
- **Pandas**: Data manipulation and time-series operations
- **NumPy**: Numerical computations
- **Scikit-learn**: Machine learning models and metrics
- **XGBoost**: Gradient boosting implementation
- **Matplotlib/Seaborn**: Data visualization

**Statistical Analysis**:
- **SciPy**: ANOVA, hypothesis testing
- **Statsmodels**: Regression analysis
- **SHAP**: Feature importance interpretation

### Data Pipeline

1. **Data Ingestion**: Load telemetry and track data
2. **Feature Engineering**: Create 20+ derived features
3. **Time-Series Quantization**: Discretize continuous signals
4. **Data Normalization**: Standardize feature scales
5. **Train/Test Split**: Temporal and circuit-based splits
6. **Model Training**: Hyperparameter optimization
7. **Evaluation**: R², RMSE, MAE metrics
8. **Interpretation**: SHAP analysis and visualizations

### Code Structure

```python
# Simplified workflow
1. load_telemetry_data()
2. engineer_track_features()
3. quantize_time_series()
4. build_driver_features()
5. train_model(features, target=lap_time)
6. evaluate_performance(test_set)
7. analyze_feature_importance()
8. visualize_predictions()
```

## Applications

### Racing Teams (F1 Constructor Applications)

**Strategy Optimization**:
- Lap time predictions for circuit planning
- Driver-circuit matching analysis
- Tire strategy simulations
- Pit stop timing optimization

**Car Development**:
- Understanding which car characteristics matter most
- Prioritizing aerodynamic vs. mechanical grip
- Optimizing for specific circuit types

### Sports Analytics

**Broadcasting & Media**:
- Real-time prediction graphics
- Performance comparison visualizations
- Insightful commentary support

**Fantasy Sports**:
- Driver performance forecasting
- Optimal team selection
- Risk assessment for picks

### Data Science Showcase

**Transferable Skills**:
- **Feature Engineering**: Complex interaction terms
- **Time-Series Analysis**: Quantization techniques
- **Statistical Rigor**: ANOVA, hypothesis testing
- **Model Interpretation**: SHAP analysis
- **Domain Expertise**: F1 racing knowledge integration

## Insights & Discoveries

### Track Length Dominance (56.40%)
- Longer tracks = longer lap times (obvious but quantified)
- Serves as baseline normalization factor
- Other features explain residual variation

### Elevation Impact (20.81% combined)
- Underestimated factor in lap time prediction
- Affects engine load and aerodynamics
- More important than raw corner count

### Curvature Complexity (9.33%)
- Standard deviation more important than mean
- Track-to-track variation matters
- Technical circuits favor skilled drivers

### Driver Differences (Verstappen vs. Norris)
- Quantified performance gaps on complex tracks
- Tire management measurably different
- Temperature/weather sensitivity varies

## Challenges & Solutions

### Challenge 1: High-Dimensional Data
**Solution**: Feature selection via ANOVA and recursive feature elimination

### Challenge 2: Non-Linear Relationships
**Solution**: Time-series quantization and interaction terms

### Challenge 3: Driver Heterogeneity
**Solution**: Driver-specific features and mixed-effects modeling

### Challenge 4: Overfitting Risk
**Solution**: Cross-validation, regularization, and temporal splits

## Future Enhancements

### Planned Additions
- **Real-time predictions**: Live race lap time forecasting
- **Strategy simulation**: Pit stop and tire strategy optimization
- **Weather integration**: Rain impact on lap times
- **Machine learning ensemble**: Combining multiple models

### Advanced Features
- Tire compound effects (soft vs. medium vs. hard)
- Fuel load degradation curves
- DRS (Drag Reduction System) impact
- Traffic and overtaking difficulty

### Deep Learning Exploration
- LSTM for sequential telemetry data
- CNN for circuit image analysis
- Transformer models for attention-based predictions

## Conclusion

This project demonstrates **end-to-end data science expertise** in a complex, real-world domain:

- **94.8% R² accuracy** through rigorous feature engineering
- **Statistical validation** via ANOVA and hypothesis testing
- **Interpretable models** with SHAP feature importance
- **Domain integration** combining F1 knowledge with ML techniques

The framework is **transferable to financial modeling** (trading strategies), **sports analytics** (performance prediction), and **any time-series regression problem** requiring sophisticated feature engineering.

### Key Takeaways
1. **Track length** is the dominant predictor (56.40%)
2. **Elevation variation** significantly impacts lap times (20.81%)
3. **Driver characteristics** create measurable performance differences
4. **Advanced feature engineering** unlocks predictive accuracy

## Repository

**GitHub**: [SidShah2953/F1-Telemetry-Analysis](https://github.com/SidShah2953/F1-Telemetry-Analysis)

**Contents**:
- Complete data pipeline code
- Feature engineering notebooks
- Statistical analysis scripts
- Model training and evaluation
- Visualization tools
- Documentation and results

---

*This project showcases the intersection of machine learning, statistical analysis, and domain expertise—skills directly applicable to quantitative finance and data-driven decision making.*

## Disclosure

The views expressed here are my own personal opinions. This is **not investment advice** and should not be relied upon as such. Nothing here is connected to, endorsed by, or written on behalf of Rosenblatt Securities.

---

Source: https://siddhants.com/projects/f1-lap-time-prediction/
About the author: https://siddhants.com/about.md
Site index for agents: https://siddhants.com/llms.txt