UNSUPERVISED LEARNING PROJECT
REPORT
1. Main Objective of the Analysis
The main objective of this analysis is to apply unsupervised learning techniques,
specifically clustering, to identify hidden patterns among Formula 1 World Championship–
winning seasons. The goal is to group similar championship performances together in order
to understand what differentiates highly dominant seasons from more competitive ones.
This analysis provides value to stakeholders such as motorsport analysts, team strategists,
historians, and media professionals by offering a structured, data-driven way to interpret
championship dominance, performance consistency, and historical trends in Formula 1.
2. Description of the Dataset
The dataset used in this project is the Formula 1 World Championship Winners dataset,
compiled from publicly available Formula 1 records and sports analytics sources. The
dataset contains season-level information for drivers who won the Formula 1 World
Championship.
The dataset includes information such as the championship year, winning driver, constructor
(team), number of race wins, total points scored, and driver nationality. Since the dataset
focuses on winners only, it is well suited for analyzing patterns among top-performing
seasons rather than overall race participation.
Key Attributes:
● Season: Year of the championship
● Driver: Name of the championship-winning driver
● Constructor: Team of the winning driver
● Wins: Number of races won in the season
● Points: Total points scored in the season
● Nationality: Driver nationality
The objective with this dataset is to uncover natural groupings among championship-winning
seasons based on performance characteristics.
3. Data Exploration and Preparation
Initial data exploration involved reviewing summary statistics and understanding the
distribution of numerical features such as wins and points across seasons. The analysis
revealed variation in season lengths and scoring systems across different Formula 1 eras.
Key data preparation steps included:
● Removing non-numeric identifiers (driver name, team, nationality) from the modeling
stage
● Handling missing or inconsistent values where applicable
● Scaling numerical features using StandardScaler to ensure fair distance-based
clustering
● Verifying that no single feature dominated the clustering process
These steps ensured the dataset was suitable for unsupervised learning and produced
stable clustering results.
4. Unsupervised Models Trained
To meet the project requirements, three variations of unsupervised learning models were
trained and compared.
Model 1: K-Means Clustering
K-Means clustering was applied using different numbers of clusters (k = 2, 3, and 4). The
Elbow Method and Silhouette Score were used to evaluate cluster separation.
● Strengths: Simple, interpretable, and effective for numeric performance data
● Limitations: Sensitive to outliers and requires predefined number of clusters
Model 2: Hierarchical Clustering
Hierarchical clustering using Ward linkage was applied to observe how championship
seasons naturally group together. A dendrogram was used to identify a meaningful number
of clusters.
● Strengths: Does not require predefining cluster count, strong for visualization and
storytelling
● Limitations: Less scalable and sensitive to noise
Model 3: DBSCAN
DBSCAN was used to identify dense regions of similar championship performances and
detect exceptionally dominant or unusual seasons as outliers.
● Strengths: Identifies noise and does not require specifying number of clusters
● Limitations: Sensitive to parameter selection and varying data density
5. Recommended Final Model
After comparing all three approaches, K-Means clustering with three clusters was
selected as the final model. This model provided the clearest separation between dominant,
moderately dominant, and competitive championship seasons.
K-Means was chosen because it produced stable clusters, was easy to interpret, and aligned
well with the main objective of grouping championship-winning seasons based on
performance intensity.
6. Key Findings and Insights
The final clustering model revealed three distinct groups of championship-winning seasons:
● Cluster 1 – Dominant Seasons: Characterized by a high number of wins and
points, often associated with technical superiority and strong team performance.
● Cluster 2 – Moderately Dominant Seasons: Seasons with solid performance but
fewer wins, indicating closer competition.
● Cluster 3 – Competitive Seasons: Championships won with relatively fewer wins,
suggesting highly competitive seasons where consistency played a major role.
The analysis also showed that certain constructors appear frequently in dominant clusters,
highlighting periods of sustained dominance in Formula 1 history.
7. Limitations and Next Steps
While the analysis provided meaningful insights, it has some limitations:
● The dataset includes only championship winners, excluding other drivers
● Changes in scoring systems and regulations across seasons affect direct
comparisons
● Limited availability of contextual factors such as team budgets or car regulations
Suggested Next Steps:
● Expand the dataset to include all drivers and constructors
● Incorporate race-level or qualifying data
● Segment analysis by Formula 1 eras to account for regulation changes
● Combine clustering results with supervised learning models for deeper insights
8. Conclusion
This project demonstrates how unsupervised learning can be used to extract valuable
insights from sports data. By clustering Formula 1 championship-winning seasons, the
analysis highlights patterns of dominance and competition across different eras.
The final model provides a clear, interpretable framework for understanding championship
performance and offers a strong foundation for further analytical exploration.