Unsupervised Learning
in Finance
An 8-week practical workshop that moves through clustering, dimensionality reduction, and anomaly detection - applied directly to financial datasets you'd actually encounter on the job.
What the program actually covers
Most finance courses treat machine learning as a theoretical appendix. This one starts with a dataset - equity return histories, macro indicators, or fund holdings - and builds outward from there. The first two weeks are deliberately slow, focused on understanding what k-means and hierarchical clustering are actually doing to your data before you tune a single hyperparameter.
Weeks three and four shift to dimensionality reduction: PCA, t-SNE, and UMAP applied to portfolio construction problems. You'll see where these methods help and where they mislead, which is a more useful lesson than a clean textbook example.
The final stretch covers anomaly detection in transaction data and time series, ending with a capstone where you bring your own dataset and apply the full pipeline under live review.
- Module 1–2: Clustering foundations - k-means, DBSCAN, hierarchical methods on return data
- Module 3–4: Dimensionality reduction - PCA and manifold methods for portfolio analysis
- Module 5: Anomaly detection - isolation forest and autoencoders on transaction series
- Module 6: Capstone - participant-led dataset project with group review session
Tools used throughout
Python, pandas, scikit-learn, matplotlib, and UMAP-learn. All open-source, no paid licences needed. Setup guide sent before week one.
Weekly schedule
| Day | Format |
|---|---|
| Tuesday | Concept session (recorded, 60 min) |
| Thursday | Live lab (90 min, small group) |
| Rolling | Assignment + written feedback |
Small cohort by design
Capped at 12 participants so the Thursday labs stay conversational. Instructor Caoimhe Ó Briain reviews each assignment individually before the next session.