I've recently been tasked with an Data Science interview assignment and looking over the variables, I wondered if it is professionally acceptable to make a new independent variable out of old ones (or at least modify the independent variables in a way to potentially get more worthwhile data).

For example, one of the variables has data on a person's birth state/territory. So instead of trying to model using 50+ categories, creating a new variable that would categorize each state by region (North East, Midwest, South, Etc.) might lead to more significant and easier to interpret analysis.

I know the objective, or what we're trying to answer, is important, but I wonder if making brand new variables or significant changes like this is appropriate for an interview assignment?



It is completely acceptable if you make out new categories out of old ones or add a different representation of them like in the case of (North East, Midwest, South, Etc.) from state this process is called feature engineering. All you need to pay attention to is that you are not loosing any information in any way when you are doing so because you want to feed your model with as much information as possible.

Sometimes it is useful to keep both the new variable you made out of other variables like for say you have a dataset containing infect_people_from_covid19 in region and population_of_region you can create a variable as ration of people infected but when building a model and feeding data you have to decide weather adding all 3 of them will be a good idea or you just need the summary variable you made.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.