Blog post

Notes on clustering experiments

2026-06-20 • clustering, anomaly detection, unsupervised learning

Clustering is often best used as an exploratory step before building a formal model. It can uncover hidden patterns and surface suspicious segments.

I found that standardizing continuous features and choosing the right distance metric had a bigger impact than the clustering algorithm itself.

In practice, I combined cluster assignments with business rules so the output was easy for analysts to review and trust.