I have a quick question: is it necessary to normalize both the dependent and independent variables for a regression task when employing ANNs? I have come across example code where normalization was applied to both, while in some cases, only the independent variables were normalized. I would greatly appreciate it if you could provide insights into the advantages and disadvantages of normalizing the dependent variable as well.

You might want to think of normalizing random variables as a change of units. If you divide 1000 m by 1000, you get 1 km (one 1000m unit). If you divide a distribution of mice weights in g by its standard deviation in g, you get a normalized unitless measurement.

In theory, a statistical model does not really care about the unit you supply to it. As long as you fulfill all the models predefined assumptions, it will work. For example, many statisticians prefer to analyze the log(x) instead of x, if log(x) better fits their models assumptions. You can do this with the dependent variable and the independent variable, or with both.

The main problem arises when you want to interpret any results. Here you have to reverse all the unit changes correctly, or your conclusions will be wrong.

This topic was automatically closed 21 days after the last reply. New replies are no longer allowed.

If you have a query related to it or one of the replies, start a new topic and refer back with a link.