Unsupervised Machine Learning Workshop

Machine Learning Bootcamp · Gabriela Granja

Two practical workshops for learning how representation, preprocessing and algorithm choice change unsupervised machine learning results. The executed notebooks are the reproducible evidence; the GitHub Pages website presents the workflows, visualisations, findings and limitations.

Workshops

1. Mushroom Workshop

An 8,124-row categorical dataset becomes 115 One-Hot encoded features after cleaning. The workshop compares PCA, t-SNE, UMAP, K-Means, agglomerative clustering, DBSCAN and HDBSCAN, then uses FP-Growth association rules to recover readable category combinations.

Key learning: visually organised embeddings do not necessarily produce clusters that agree with the reference labels. Association rules offered the clearest feature-level interpretation, while clustering results changed materially with representation and density parameters.

Read the Mushroom Workshop

2. Credit Card Workshop

A numerical behavioural dataset contains 8,950 customers and 17 modelling variables after identifier removal. The workflow covers median imputation, StandardScaler, PCA, K-Means model selection, hierarchical clustering, DBSCAN, Gaussian Mixture, customer profiling and Isolation Forest.

Key learning: three K-Means profiles provide the clearest baseline combination of internal metrics and interpretability. Engineered KPIs and log1p produce a more balanced but materially different segmentation, while UMAP/HDBSCAN reveals an alternative six-cluster structure.

Read the Credit Card Workshop

Reproducible notebooks

Workshop	Baseline	Advanced
Mushroom	workshop-clustering-Mushrooms.ipynb	workshop-advanced-unsupervised.ipynb
Credit card	workshop-clustering-creditcard.ipynb	workshop-clustering-creditcard-advanced.ipynb

Techniques applied

• Exploratory data analysis, missing-value treatment and feature scaling
• One-Hot Encoding, engineered ratios and log1p representation
• PCA, t-SNE and UMAP
• K-Means, hierarchical clustering, DBSCAN, Gaussian Mixture and HDBSCAN
• Silhouette, Calinski-Harabasz, Davies-Bouldin, ARI and NMI evaluation
• Isolation Forest anomaly exploration
• FP-Growth and association rule mining
• Sub-sampling stability analysis

Repository structure

.
├── data/
├── docs/
│   ├── assets/
│   └── figures/
├── scripts/
├── workshop-clustering-Mushrooms.ipynb
├── workshop-advanced-unsupervised.ipynb
├── workshop-clustering-creditcard.ipynb
├── workshop-clustering-creditcard-advanced.ipynb
├── mkdocs.yml
├── requirements.txt
├── requirements-docs.txt
└── README.md

Run locally

git clone https://github.com/Bootcamp-IA-MAD-P7/unsupervised-ml-workshop.git
cd unsupervised-ml-workshop
python -m venv .venv

Activate the environment on Windows:

.venv\Scripts\activate

Or on Linux/macOS:

source .venv/bin/activate

Install dependencies and launch Jupyter:

pip install -r requirements.txt
jupyter notebook

Build the documentation:

pip install -r requirements-docs.txt
mkdocs build --strict

Author

Gabriela Granja

Digital Marketing professional transitioning into Artificial Intelligence, Data Analytics and Machine Learning through hands-on technical projects.