Formula 1 Race Winner Prediction Model

A comprehensive machine learning project that predicts Formula 1 race winners using historical data from 1950 to 2024. This system compares multiple ML algorithms, automatically selects the best performer, and deploys it via a premium web application with a modern dark-mode interface.

📋 Table of Contents

• Overview
• Critical Implementation Fixes
• Dataset
• Installation
• Usage
• Features
• Model Performance
• Web Application

🎯 Overview

This project trains and compares multiple machine learning algorithms to predict Formula 1 race winners with high accuracy. The pipeline automatically selects the best-performing model and deploys it through a sleek, interactive web application.

Key Highlights

• Data Leakage Prevention: Uses pre-race cumulative_points instead of post-race points to ensure realistic results.
• Multi-Model Comparison: Evaluates Random Forest, XGBoost, Logistic Regression, Gradient Boosting, and more.
• Automatic Model Selection: Identifies the best performer using ROC-AUC and F1-Score metrics.
• Robust Preprocessing: Standardized feature scaling (StandardScaler) and automated categorical encoding.
• Premium Web Interface: Modern Flask-powered web app with F1-themed dark mode and dynamic circuit metadata.
• Real-Time Predictions: Instant winning probability for any driver/circuit/team combination.

⚠️ Critical Implementation Fixes

The model includes several critical improvements over traditional F1 predictors:

1. Fixing Data Leakage: Replaced post-race points with cumulative_points (points earned before the current race). This ensures the model only uses information that would be available prior to the race start.
2. Feature Scaling: Implemented StandardScaler to normalize features like cumulative_points (0-400+) and grid_position (1-20), preventing large-scale variables from dominating the model.
3. Class Imbalance Handling: Applied class_weight='balanced' and optimized for Weighted F1-Score to account for the fact that only one driver wins per race.

📊 Dataset

The dataset is sourced from the Kaggle Formula 1 World Championship (1950-2024), compiled from the Ergast API.

The pipeline integrates 14 CSV files:

1. circuits.csv: Circuit metadata (location, country, coordinates).
2. constructor_results.csv: Constructor race points.
3. constructor_standings.csv: Constructor championship positions.
4. constructors.csv: Team names and nationalities.
5. driver_standings.csv: Driver championship points and wins.
6. drivers.csv: Driver names, nationalities, and DOB.
7. lap_times.csv: Lap-by-lap timing data.
8. pit_stops.csv: Pit stop durations and lap numbers.
9. qualifying.csv: Q1, Q2, and Q3 session times.
10. races.csv: Race calendar and metadata.
11. results.csv: Final race results (primary target data).
12. seasons.csv: Historical season links.
13. sprint_results.csv: Sprint race outcome data.
14. status.csv: Finish status codes (Finished, DNF, etc.).

🚀 Installation

1. Clone the repository

   git clone https://github.com/heyisula/f1.git
   cd f1

2. Install dependencies

   pip install -r requirements.txt

🚀 Usage

Step 1: Train the Model

Open and run all cells in train.ipynb:

 jupyter notebook train.ipynb

The notebook will:

• Load and preprocess data from the data/ directory.
• Implement data leakage fixes and feature engineering.
• Train multiple models (XGBoost, RandomForest, Gradient Boosting).
• Save the best model and preprocessors to out/models/f1_model_data.pkl.

Step 2: Launch the Web App

Start the Flask server:

python app.py

Open your browser and navigate to http://127.0.0.1:5000. You can:

• Select a driver, team, and circuit.
• Watch as the Laps field auto-fills based on the circuit selected.
• Get an AI-predicted winning probability.

🔧 Features

Current Feature Set

Feature	Description	Importance
Grid Position	Starting position on the grid (1-20)	High
Cumulative Points	Points earned entering the race	High
Circuit ID	Encoded circuit identifier	Medium
Driver/Team ID	Encoded driver and constructor identifiers	Medium-High
Driver Age	Calculated age at race time	Low
Laps	Total race distance	Low

Potential Future Features

• Qualifying gap to pole (seconds)
• Reliability index (DNF rate over last 5 races)
• Teammate vs. Teammate historical performance
• Weather conditions (Rain/Dry probability)

📊 Model Performance

Metric	Target	Interpretation
Weighted F1-Score	~0.96	High overall classification accuracy
ROC-AUC	~0.95	Excellent model discriminative ability
Winner Precision	~0.55-0.65	Realistic given single-winner probability

Note: Metrics represent the optimized Gradient Boosting model.

📄 License

This project is open source and available under the MIT License.

🙏 Acknowledgments

• Dataset: Ergast F1 API via Kaggle
• Inspiration: F1 racing analytics community