Description
Turns fitted Gaussian mixture components into meaningful behavioural archetypes by profiling means, covariances and membership uncertainty against qualitative evidence.
A GMM outputs, for each component, a mixing weight, a mean vector and a covariance matrix, plus each user's posterior membership probabilities. Interpretation translates these into archetypes: which features deviate most from the population mean (standardised differences), how variables co-vary within a component, how large and how certain each group is, and how components differ on outcomes not used in clustering. Careful interpretation keeps the probabilistic nature visible, reporting uncertain users and overlap rather than turning components into crisp personas. It avoids naming clusters from a single feature and validates names with qualitative review of exemplar users.
Attributes
- Discipline
- Computational & Machine Learning
- Loss type
- Latent
- Goal
- Segment users, Synthesise data
- Stage
- Definition
- Data
- Mixed
- Effort
- Days
Steps
- For each component, compute standardised mean differences from the overall mean for every feature and rank the largest.
- Inspect within-component covariance to see characteristic co-variation (e.g. high trade frequency with small order size).
- Report mixing weights and the distribution of posterior probabilities, identifying users with maximum membership below 0.7.
- Profile components on external variables not used in clustering, such as retention, revenue and survey attitudes.
- Pull 5-10 high-probability exemplar users per component for session replay or interviews and write a draft archetype description.
- Name archetypes by behaviour, not demographics, and record the evidence for each name.
Sources
- Cluster Analysis for Researchers (Charles Romesburg)
- Pattern Recognition and Machine Learning (Christopher Bishop)
- Methods Atlas (Vivek Gopinath)
- Cluster Analysis for Researchers
- Pattern Recognition & ML (Bishop)
- Methods Atlas (Gopinath)
- The Inquiry Series, Ch 51 — Behavioral Clustering & Probabilistic Segmentation
Also known as
- Mixture component profiling
- Archetype interpretation
Categories
- Analytics & Behaviour
- Computational & Machine Learning