Skip to content

Gaussian Mixture Models

Soft clustering with probability distributions: how GMMs work, how they're fitted and when to use them.

Editorial team 1 min read

A Gaussian mixture model (GMM) assumes data comes from a mix of several normal distributions, each representing a cluster.

Soft Assignments

Unlike k-means, which assigns each point to one cluster, a GMM gives the probability that each point belongs to each cluster. Useful when groups overlap.

Fitting

GMMs are usually fitted with the expectation-maximisation (EM) algorithm:

  1. E-step: estimate cluster membership probabilities.
  2. M-step: update each component's mean, covariance and weight.
  3. Repeat until stable.

Covariance Types

Components can be spherical, diagonal or fully flexible ellipses. More flexible shapes fit better but need more data.

Choosing the Number of Components

Use criteria such as BIC or AIC, which balance fit against complexity.

Uses

  • Clustering with overlapping groups.
  • Density estimation.
  • Anomaly detection: points with low likelihood are unusual.

Limitations

Sensitive to initialisation; can converge to poor solutions. Run several initialisations and assume roughly elliptical clusters.

More in Machine learning

All Machine learning guides →
Machine learning Guide · 2 min

Linear Regression Explained

The simplest predictive model: how linear regression fits a line through data, how to read its coefficients, and when it breaks down.

Machine learning 2 min read 17 Sep 2026

Machine learning Guide · 2 min

Logistic Regression for Classification

Despite its name, logistic regression is a classification method. How it produces probabilities and why it remains a strong baseline.

Machine learning 2 min read 16 Sep 2026

Machine learning Guide · 2 min

Decision Trees

How decision trees split data with simple questions, why they are easy to explain, and why single trees overfit.

Machine learning 2 min read 15 Sep 2026

Machine learning Guide · 2 min

Random Forests

Why averaging many randomised decision trees produces a robust, accurate model with little tuning.

Machine learning 2 min read 14 Sep 2026