F1 2026 Drivers' Championship Prediction

A machine learning project that predicts the 2026 Formula 1 Drivers' Championship using historical race data loaded via FastF1 and a linear regression model — with an interactive Streamlit dashboard for exploring predictions, model performance, and driver deep-dives.

---

Table of Contents

• Project Overview
• Results
• Screenshots
• Project Structure
• Setup & Installation
• Usage
• Streamlit Dashboard
• Model Details
• Tech Stack

---

Project Overview

Goal: Predict which driver will win the 2026 F1 Drivers' Championship based on historical performance features derived from 2023–2025 season data.

Approach:

1. Load historical F1 race results (2023–2025) via FastF1 API
2. Engineer per-driver performance features (win rate, podium rate, DNF rate, average finish, rolling averages, etc.)
3. Train a linear regression model on historical championship points
4. Predict 2026 championship points and rank all drivers

---

Results

🏆 2026 Predicted Champion: Max Verstappen

Pos	Driver	2026 Team	Predicted Points	Win Probability
1	Max Verstappen	Red Bull Racing	400	29.9%
2	Lando Norris	McLaren	343	16.9%
3	Charles Leclerc	Ferrari	328	14.5%
4	Carlos Sainz	Williams †	277	8.7%
5	Oscar Piastri	McLaren	262	7.5%
6	George Russell	Mercedes	226	5.2%
7	Lewis Hamilton	Ferrari †	206	4.3%
8	Sergio Perez	(Red Bull) ‡	140	2.2%
9	Fernando Alonso	Aston Martin	67	1.1%
10	Oliver Bearman	Haas F1 Team	50	0.9%

Notable 2026 driver moves (not reflected in training data):

Driver	Previous Team	2026 Team
Lewis Hamilton	Mercedes	Ferrari
Carlos Sainz	Ferrari	Williams
Kimi Antonelli	— (rookie)	Mercedes
Liam Lawson	RB	Red Bull Racing
Esteban Ocon	Alpine	Haas F1 Team
Nico Hulkenberg	Haas F1 Team	Kick Sauber / Audi
Jack Doohan	— (reserve)	Alpine
Franco Colapinto	Williams	Alpine

† Team label corrected to 2026 reality; model predictions are based on driver performance history (2023–2025) at their previous constructors.
‡ Sergio Perez was replaced at Red Bull by Liam Lawson for 2026; his prediction reflects historical Red Bull-era performance.

Limitation: The model predicts based on a driver's historical performance regardless of constructor change. Actual results may differ significantly when a driver switches to a weaker or stronger team.

Model Performance

Metric	Train	Test
R²	0.9995	0.9750
RMSE	2.89 pts	14.66 pts
MAE	—	7.43 pts
Cross-val R² (mean ± std)	0.9935 ± 0.003	—

The model achieves an R² of 0.975 on the test set, indicating strong predictive accuracy. The high training R² (0.9995) with good test performance confirms the model generalises well without significant overfitting.

Top Feature Importances

Feature	Coefficient	Direction
RollingPodiumRate	−70.49	Negative
PodiumRate	+59.21	Positive
AvgPoints	+23.94	Positive
DNFRate	+16.71	Positive
WinRate	−7.72	Negative

Note: Negative coefficients for rolling features vs positive for season-aggregate features reflect multi-collinearity in the linear model; the net effect of podium/win performance is strongly positive.

---

Screenshots

Championship Predictions — Top 10

Win Probability Distribution

Model Performance: Actual vs Predicted

Feature Importance

---

Project Structure

F1/
├── app.py                          # Streamlit interactive dashboard
├── data/
│   ├── raw/                        # Raw FastF1 cached data
│   ├── processed/
│   │   └── driver_features.csv     # Engineered feature matrix
│   └── predictions/
│       ├── 2026_championship_predictions.csv
│       └── prediction_summary.txt
├── docs/
│   └── images/                     # README screenshots
├── models/
│   ├── feature_importance.csv      # Model coefficients per feature
│   └── model_metrics.csv           # Train/test performance metrics
├── notebooks/
│   ├── 01_data_exploration.ipynb
│   ├── 02_feature_engineering.ipynb
│   ├── 03_model_training.ipynb
│   └── 04_predictions.ipynb
├── src/
│   ├── data/
│   │   └── load_fastf1.py          # FastF1 data loading utilities
│   ├── features/
│   │   └── build_features.py       # Feature engineering pipeline
│   ├── models/
│   │   ├── train_model.py          # Model training script
│   │   └── predict.py              # Prediction script
│   └── visualization/
│       └── visualize.py            # Plotting utilities
├── requirements.txt
└── README.md

---

Setup & Installation

Prerequisites

• Python 3.9+
• pip

1. Clone the repository

git clone https://github.com/isthatpaul/F1.git
cd F1

2. Create and activate a virtual environment

# Windows
python -m venv .venv
.venv\Scripts\activate

# macOS / Linux
python -m venv .venv
source .venv/bin/activate

3. Install dependencies

pip install -r requirements.txt

---

Usage

Option A — Jupyter Notebooks (recommended)

Run the four notebooks in order:

Notebook	Purpose
notebooks/01_data_exploration.ipynb	Explore raw FastF1 data
notebooks/02_feature_engineering.ipynb	Build the driver feature matrix
notebooks/03_model_training.ipynb	Train & evaluate the linear regression model
notebooks/04_predictions.ipynb	Generate 2026 championship predictions

jupyter lab

Option B — Python scripts

# 1. Load and process data
python src/data/load_fastf1.py

# 2. Build features
python src/features/build_features.py

# 3. Train model
python src/models/train_model.py

# 4. Generate predictions
python src/models/predict.py

Predictions are saved to data/predictions/2026_championship_predictions.csv.

---

Streamlit Dashboard

An interactive web dashboard (app.py) provides three views:

Page	Description
🏆 Championship Predictions	Bar chart of predicted points, win probability pie chart, and full standings table
📊 Model Performance	Train vs test metrics (R², RMSE, MAE) and feature importance chart
🔍 Driver Deep Dive	Per-driver breakdown with radar/detail charts

Run the dashboard

streamlit run app.py

The app opens at https://pauls-f1-lab.streamlit.app/ by default.

---

Model Details

Algorithm: Linear Regression (scikit-learn LinearRegression)

Target variable: Season championship points

Input features:

Feature	Description
AvgPoints	Average points per race in the season
AvgFinish	Average finishing position
AvgGrid	Average qualifying grid position
WinRate	Fraction of races won
PodiumRate	Fraction of races with a podium finish
DNFRate	Fraction of races not finished
AvgPositionChange	Average positions gained/lost vs grid
PrevYearPoints	Championship points from the previous season
PrevYearPodiums	Podium count from the previous season
RollingAvgPoints	Rolling 3-season average points
RollingPodiumRate	Rolling 3-season average podium rate

Training data: 2023–2025 F1 seasons

---

Tech Stack

Library	Purpose
FastF1	F1 race data API
scikit-learn	Machine learning
pandas	Data manipulation
numpy	Numerical computing
matplotlib	Static visualisation
seaborn	Statistical visualisation
plotly	Interactive charts
Streamlit	Interactive web dashboard

---

Predictions are based on historical data and statistical modelling — not a guarantee of future results.