# Expertise: Modeling

To understand your data it is useful, perhaps critical, to model its structure. Models come in many flavors.

**Statistical Models** are the most common. They are best used when for some reason you believe that your data has characteristics of a certain type. For example, if you want a model of how many events might occur in a window of time, you use a Poisson process. If you want a model of a random variable without knowing its true underlying distribution, then you might use an ordinary Gaussian, or perhaps a Gaussian mixture.

**Geometric Models** are useful when you want your data to speak for itself. These models determine the most likely shape of a dataset and codify points by their projections onto these shapes. For example, if your dataset lies in a ring, the circle is a good geometric model. How close a point is to the circle provides a new way to think of normal or outlying data-points. More generally, a given dataset can have many geometric models, each providing a different view of its shape.

GDA has extensive expertise with both types of models. In fact, we are particularly good at using them together to make the best possible use of the information that is available about the data before analysis.