19. Classification


This chapter is under development. When it’s finished, this note will be removed.

Thus far we have studied models for regression, the process of making continuous, numerical estimations based on data. We now turn our attention to classification, the process of making categorical predictions based on data. For example, weather stations are interested in predicting whether tomorrow will be rainy or not using the weather conditions today.

Together, regression and classification compose the primary approaches for supervised learning, the general task of learning a model based on observed input-output pairs.

We may reconstruct classification as a type of regression problem. Instead of creating a model to predict an arbitrary number, we create a model to predict a probability that a data point belongs to a category. This allows us to reuse the machinery of linear regression for a regression on probabilities: logistic regression.