Scala and Spark for Big Data Analytics

How does a Gaussian mixture model work?

Using GMM is a popular technique of soft clustering. GMM tries to model all the data points as a finite mixture of Gaussian distributions; the probability that each point belongs to each cluster is computed along with the cluster related statistics and represents an amalgamate distribution: where all the points are derived from one of K Gaussian subdistributions having own probability. In short, the functionality of GMM can be described in a three-steps pseudocode:

1. Objective function: Compute and maximize the log-likelihood using expectation–maximization (EM) as a framework
2. EM algorithm:
   • E step: Compute the posterior probability of membership -i.e. nearer data points
   • M step: Optimize the parameters. ...