Model Description
This model uses K-Means clustering to group wholesale customers based on annual spending patterns.
It identifies patterns across six numerical features from the Wholesale Customers Dataset:
Intended Uses & Limitations
This model is intended only for assignment and analytical use in the context of CSC310 coursework.
It is not designed for production systems or real world deployment.
Intended Uses
- Demonstrate how to apply K-Means to tabular data.
- Show how to share models with documentation using Hugging Face.
- Provide an example of an unsupervised machine learning workflow.
Limitations
- Dataset is small and not representative of bigger consumer populations.
- Right-skewed data, and outliers.
- K-Means may not reflect real world data.
- Clusters are statistical groupings, not value based categories.
How to Get Started with the Model
You can download and test this model or notebook directly from this repository.
from huggingface_hub import hf_hub_download
import joblib, pandas as pd
Download trained model
hf_hub_download(repo_id="CSC310-fall25"/wholesale_clustering", filename="model.joblib")
Hyperparameters
| Hyperparameter | Value |
|---|
| algorithm | lloyd |
| copy_x | True |
| init | k-means++ |
| max_iter | 300 |
| n_clusters | 3 |
| n_init | auto |
| random_state | 42 |
| tol | 0.0001 |
| verbose | 0 |
Model Plot
km3 = cluster.KMeans(n_clusters = 3, random_state = 42)
km3
Load model
model = joblib.load("model.joblib")
Model Card Authors
Name: Steven Doss
Model Card Contact
Citation
Evaluation Results
Silhouette score is 0.46 which confrims a vaid level of separation between customers. Which means that there is enough diversity in spending across the clusters that customers are similar within a group to be different form other clusters.,
2.
Adjusted Mutual Informatio - This score is low. (0.10). However, it is expected because KMeans found 3 spending groups while labels are only defined by two (Horeca & Retail).
