Splinter Segmentation Engine

🧠 How K-Means Clustering Works

K-Means groups users into clusters by similarity across multiple features. Here's what happened:

1. Feature extraction — Each user is represented as a vector of normalised features: LTV, trades, days since last trade, app sessions, funded status, risk score, and account age
2. K-Means++ initialisation — Smart centroid seeding to avoid poor starting positions
3. Tested k=3 to k=7 — Ran clustering for each k value
4. Silhouette scoring — Measures how well each user fits its cluster vs neighbouring clusters (score: -1 to 1, higher = better separation)
5. Best k selected — Chose the k with highest silhouette score
6. Auto-naming — Each cluster named based on its dominant traits (LTV, activity, asset mix)