Mixture-Based Classification
Luca Scrucca, Chris Fraley, Thomas Brendan Murphy, Adrian E. Raftery · 2023
Classification is an instance of supervised learning, where the class of each observation is known. Unlike in the unsupervised case, the main objective here is to build a classifier for classifying future observations. This chapter describes probabilistic classification following a mixture-based approach. It describes various Gaussian mixture models for supervised learning. The implementation available in mclust is presented using several data analysis examples. Different ways of assessing classifier performance are also discussed. The problem of unequal costs of misclassification and the classification with unbalanced classes is presented, followed by solutions implemented in mclust. The chapter concludes with an introduction to the semi-supervised classification problems, in which only some of the training data have known labels.