How to present categorical variables in ANN

Categorical variables (non-numeric variables) are often hard to use in Artificial Neural Network (ANN ) or other data mining tools. To be perfectly clear, categorical variables are a finite set of values which are not in an order. For example, linguistic terms like low and high or Countries and States are categorical variables. Usually, data mining models cannot work with non continuous numbers. On the other hand, there are some cases that using this type of variables is inevitable. A few ways exist to present categorical variables in ANN or other data mining models. The simplest way to present these variables is using dummy variables. It means, a single categorical variable with n values is converted to the n new dummy variables. For example, if you have a field “state” with 50 text labels, you should create 50 different new variables with value of 0 and 1. If a record has the value "AL" in the variable State, the new dummy column representing "AL" will have value "1", and all 49 other state dummy columns will have value "0". The following example illustrates this process for 3 states.

State               AL         GA           FL

Alabama          1             0              0

Georgia           0             1              0

Florida            0             0              1

The major disadvantage of this method is that it expands the number of input variable dramatically and this can affect the model sensitivity. Some software handles this process automatically. To reduce the number of dummy variables, classes with small representation can be presented under an "Other" category.  Although this discards information distinguishing those small categories from one another, it can potentially reduce the number of inputs substantially.

References:

http://abbottanalytics.blogspot.com/2005/04/beware-of-automatic-handling-of.html

https://www.mathworks.com/support/solutions/en/data/1-8H0STM/index.html?product=SL&solution=1-8H0STM