Cat Breed Analysis

Methodology

Data Collection

We queried a Large Language Model (LLM) to obtain ratings for various traits across many different cat breeds. This step can also be done via scaping data from some cat database website such as CFA or TICA.

Dimensionality Reduction

Given the high-dimensional nature of our trait data, we applied Principal Component Analysis (PCA) to reduce the data to 2 dimensions. This allows us to visualize the relationships between cat breeds while preserving the most important variance in the data.

The two principal components that we plot are defined as linear combinations of the original traits:

PC1 (Overall Energy Level):

PC1 = 0.183 × Affection + 0.538 × Playfulness + 0.481 × Energy + 0.582 × Vocalness + 0.328 × Intelligence

PC2 (Overall Affectionate):

PC2 = 0.588 × Affection - 0.139 × Playfulness - 0.555 × Energy + 0.527 × Vocalness - 0.221 × Intelligence

These components explain the maximum variance in the original 5-dimensional trait space, allowing us to visualize breed relationships in 2D while preserving the most important information.

Clustering Analysis

We used the K-means clustering algorithm to identify natural groupings of cat breeds based on the two principle traits that we have indentified in the previous step. To determine the optimal number of clusters, we employed the Within-Cluster Sum of Squares (WCSS) method, also known as the elbow method.

WCSS plot showing the elbow at k=4, indicating the optimal number of clusters

The WCSS plot above clearly shows an "elbow" at k=4, confirming that 4 clusters provide the optimal balance between model complexity and explanatory power for our cat breed dataset.

Source code can be found here

Conclusion

Our analysis successfully identified 4 distinct clusters of cat breeds based on their overall energy levels and related characteristics:

High Energy, High Affectionate Cluster: Breeds that are both very active, playful, and very affectionate and very intelligent.

Breeds: Siamese, Burmese, Sphynx, Oriental Shorthair, Balinese, Devon Rex, Tonkinese, Oriental Longhair
High Energy, Moderate Affectionate Cluster: Breeds that are very active, energetic but not as affectionate and not as intelligent.

Breeds: Bengal, Abyssinian, Cornish Rex, Egyptian Mau, Japanese Bobtail, Ocicat, Savannah, Somali, Chausie
Moderate Energy Cluster: Breeds that are relatively calm, they are moderate in both energy and affectionate levels.

Breeds: Maine Coon, Turkish Van, Havana Brown, LaPerm, Pixiebob, Siberian, Singapura, Snowshoe, Turkish Angora, Khao Manee, Korat, Munchkin
Low Energy Cluster: Breeds that are generally sedentary, preferring rest and gentle interaction. Within this cluster some can be more affectionate than others.

Breeds: Ragdoll, Scottish Fold, British Shorthair, Norwegian Forest, Russian Blue, Manx, Ragamuffin, Selkirk Rex, American Shorthair, British Longhair, Cymric, Nebelung

Implications

It is somewhat surprising that affectionate and energy are not independent criteria for the clustering: the naive approach to making 4 clusters would be high/low energy + high/low affectionate. Instead, energy level seems to be a more prominent feature.

This clustering analysis provides valuable insights for perspective cat parents so that they can make informed decisions about which breeds might be the best fit for their living situation, activity level, and lifestyle preferences. If they like one breed which is not available, they can find the similar breeds using our clustering.