Gaussian Mixture Model Clustering

Each point belongs to every cluster, but has a different level of membership. It assumes that each cluster has a certain statistical distribution.

Expectation - Maximization Algorithm

Advantages :

• Soft Clustering, sample members of multiple clusters.
• Cluster shape flexibility. (cluster can contain another cluster inside of it)

Disadvantages :

• Sensitive to initialization values
• Slow convergence rate
• Possible to converge to a local optimum

Process:

1. Initialize k Gaussian distributions
2. Soft cluster the data into the Gaussian distributions (Expectation step)
3. Re-estimate the Gaussian (Maximization step)
4. Evaluate log-likelihood to check for convergence
5. Repeat from Step.2 until convergence