Summary

This project analyzes Formula 1 race tracks using unsupervised machine learning techniques, aiming to identify meaningful clusters based on physical and performance-related characteristics. By applying K-Means clustering, we group circuits into distinct types, helping fans and analysts better understand strategic and technical demands across the championship calendar. The result is a data-driven classification of tracks into three main categories: "Modern All-Rounders", "Classic Speed Circuits", and the unique "Technical Street Challenge".

Introduction

Formula 1 circuits vary widely in design, surface, and flow. These differences significantly impact car performance, race strategies, tire choices, and driver behavior. Instead of treating each circuit as a standalone case, this project uses clustering to group similar tracks together based on their structural features and driving characteristics.

Project Scope and Rationale

The goal of this project is to classify and cluster F1 circuits using only track-specific data that does not rely on driver behavior or team strategies. For this reason, the dataset prioritizes features that describe the layout and physical attributes of each track. These include:

• Asphalt features: Abrasion

• Layout complexity: Number of corners, corner density

• Composition of corner types: High-, Medium-, and Low-speed corners

The central idea is to maintain independence from external variables like weather, team updates, or driver skill, focusing instead on what makes each circuit unique from a structural perspective.

How the Clustering Was Performed

The clustering methodology followed a structured, business-agnostic yet analytically sound approach designed to extract clarity from complexity.

We used K-Means Clustering, a well-established unsupervised learning technique that groups data points into a specified number (k) of mutually exclusive categories. Each group is formed to minimize internal variation, which means that tracks within the same cluster share similar performance-defining features.

K-Means was chosen for several reasons:

• It performs particularly well with standardized, continuous numerical features such as the ones used here.

• It is interpretable and repeatable, creating clearly separated groups that can be easily explained and analyzed.

• It offers straightforward implementation and visual inspection, enabling rapid testing of different assumptions and configurations.

Before clustering, all features were standardized to ensure equal weight and to prevent metrics with large values from dominating the outcome.

We then applied K-Means with a range of values for k (2 to 9), evaluating each with two key diagnostics:

• The elbow method, which tracks the reduction in clustering error (inertia) as more groups are added.

• The silhouette score, which measures how well-defined each cluster is. A higher score means better-defined and more cohesive clusters.

While the silhouette score peaked at k=2, the elbow method showed diminishing returns after k=3. Choosing k=3 provided a practical balance between statistical performance and analytical usefulness. It also allowed us to preserve the distinctiveness of Circuit de Monaco, which otherwise gets grouped inappropriately.

Details on how the clustering was performed can be found at my Github page, at this repo: https://github.com/cld05/F1_tracks_clustering 

Expectation from Features

Each feature in the analysis is selected for its relevance to real-world circuit dynamics:

• Abrasion is expected to differentiate tracks with high tire wear (e.g., Spa, Monza) from smoother surfaces (e.g., Monaco).

• Corner Density is expected to identify tight, technical tracks with frequent direction changes.

• Corner Entropy would highlight whether tracks offer a diverse challenge or are biased toward certain corner types.

• Flow Score, refined through multiple iterations, is designed to capture the overall rhythm of the circuit—tracks with long straights and flowing sections versus those with constant stops and starts.

Results and Interpretation

Using k=3, the clustering resulted in the groups described in the following subsections.

Conclusion

This clustering analysis of Formula 1 circuits provides a structured, quantitative view of track characteristics:

• Three distinct types of circuits were identified

• Monaco’s uniqueness was confirmed both statistically and visually

• This model offers useful insight for fans, engineers, and strategists alike

Future work could explore driver performance across clusters or compare seasonal results by cluster type. Clustering also has potential for tire strategy predictions and car setup optimization frameworks.