Gaussian Naive Bayes / by Siobhán Cronin

How it works (in a nutshell)
A family of probabilistic classifiers in which the features are assumed to be conditionally-independent (their joint distribution equals the product of their marginals). What's gaussian about it? GNB uses gaussian distributions to model features with continuous data. 

Applications

  • Spam filtering
  • Face recognition
  • Document classification (i.e. machine learning research, poetry, anthropology)

Perks

  • Can perform well in small datasets
  • Can be used when we know the relative ordering of the probabilities in question, but perhaps not the probabilities themselves
  • Simple complexity O(classes X features)

Drawbacks

  • The conditionally independent assumption may mask patterns associated with co-varying features. 

Resources

On the Optimality of the Simple Bayesian Classifier under Zero-One Loss