CS7.403b: Statistical Methods in AI

Final Project: AI For Formula One

This machine learning project predicts overtakes in Formula 1 races by analyzing telemetry and event data. Using an XGBoost classifier, we can predict whether an overtake will occur based on various race conditions and car performance metrics.

Project Overview

Formula 1 overtaking is influenced by numerous factors including car performance, track characteristics, tire conditions, and driver skill. This model aims to quantify these relationships and provide predictive insights that could be valuable for race strategy planning.

Features

• Data Preprocessing: Clean and normalize raw F1 telemetry data
• Feature Engineering: Extract and transform relevant features for optimal model performance
• Model Training: Train and optimize an XGBoost classifier for overtake prediction
• Performance Evaluation: Comprehensive metrics including accuracy, precision, recall, and F1-score
• Prediction System: Make predictions on new race scenarios

Model Weights and Dataset

Download the model weights and datasets from: Google Drive - F1 Overtake Prediction Files

Installation

Prerequisites

• Python 3.8 or higher
• pip package manager

Setup

1. Clone this repository:

   git clone git@github.com:aashreyj/f1-overtaking-prediction.git
   cd f1-overtake-prediction

2. Install dependencies:

   pip install -r requirements.txt

3. Download the model weights and datasets from: Google Drive - F1 Overtake Prediction Files

Usage

Running the Project

Exploration and Experimentation

1. Start Jupyter Lab or Notebook:

   jupyter lab

   or

   jupyter notebook

2. Open experiments.ipynb to explore the data and experiment with different approaches.

Data Preprocessing

Process the raw data using the preprocessing scripts:

python preprocessing/process_dataset_overtakes.py
python preprocessing/process_dataset_not_overtakes.py

Training

Train the model using:

python train.py

Evaluation

Evaluate the model's performance:

python eval.py

Inference

Make predictions on new data:

python infer.py

Making Predictions

You can make predictions in this ways:

1. Using the infer.py script:

python infer.py --input your_input_data.csv --output predictions.csv

Dataset Description

The project uses two main processed datasets:

• processed_data_overtakes.csv: Contains data points where overtakes occurred
• processed_data_no_overtakes.csv: Contains data points where no overtakes occurred

Data Visualizations

The resources/ directory contains several visualizations to help understand the data:

• correlation_matrix.png: Shows feature correlations
• overtakes_by_circuit.png: Displays the distribution of overtakes across different circuits
• top_circuits_for_overtaking.png: Highlights the circuits with the most overtaking opportunities
• ham_all_laps.png and ham_lap_9.png: Visualizations of Lewis Hamilton's racing data