Summary

In this chapter, we learned about mixture models, a type of hybrid model useful to solve a large collection of problems. Creating a finite mixture model is a relatively easy task given what we have learned from previous chapters. A very handy application of this type of model is dealing with an excess of zeros in count data or for example to expand a Poisson model if we observe over-dispersion. Another application we explored was about extending logistic regression to handle outliers. We also briefly discussed the central elements of performing Bayesian (or model-based) clustering. Lastly, we presented a more theoretical discussion about continuous mixture models and how these types of models are connected to concepts we already learned in previous chapters, such as hierarchical models and the Student's t-distribution for robust models.