Categorical Variables

A categorical variable is one whose values can be represented in different categories. Examples are colours of a ball, breed of dogs, and zip codes. Mapping these categorical variables in a single dimension creates a sort of dependence on each other, which is incorrect. Even though these categorical variables do not have an order or dependence, inputting them to a neural network as a single feature makes the neural network create dependence between these variables depending on the order, whereas in reality, the order does not mean anything. In this section, we will learn about the ways in which can fix this issue and train effective models.

One-hot Encoding

The easiest and the most widely used method of mapping categorical variables is to use one-hot encoding. Using this method, we convert a categorical feature into features equal to the number of categories in the feature.

Figure 5.31: Categorical feature conversion

Use the following steps to convert...