Becoming a Data Head

CHAPTER 10Understand the Classification Model

A Machine Learning algorithm walks into a bar. The bartender asks, “What'll you have?” The algorithm says, “What's everyone else having?”

Chet Hasse (@chethaase)

In the previous chapter, we described supervised learning with regression models. Specifically, regression models allow us to predict values—that is, a number, like sales—by fitting a model to our feature set. But what if you were attempting to predict a specific outcome—like will a person with a certain set of demographic features want to buy a book about data? Indeed, if you've ever wondered how companies predict if you'll click a specific ad, buy a product (and which product you'll buy), default on your car loan, get a job interview, or develop a disease, this chapter will be your guide.

Such problems—where there is a categorical variable (i.e., a label) to be predicted—employ classification models.

INTRODUCTION TO CLASSIFICATION

Classification models can predict two outcomes, called binary classification; or multiple classes, called multiclass classification.¹ Predicting if someone defaults on a car loan is binary (yes/no), while predicting which car someone will buy is multiclass (Honda, Toyota, Ford, etc.). For simplicity, we'll focus on binary classification problems. However, additional classes are a natural extension of the topics we'll discuss.

The outcomes of some classification models are often described as being “positive” or “negative.” Remember, science ...