0
$\begingroup$

I am a ml novice, though I have an extensive computing background. I am about to start a ml project, and there is something that I can't quite get my head around. If, for example, I am trying to predict mortalities of a population using a ml model can I include as a feature the count of that population that is effectively the start count less all the mortalities? Can the features be used that are derived from the target feature? A similar example, predicting the mean mass of a population at time t. Can I use the mean mass at time t-1 as an imput for the model? Another way of explaining it, can I use the column that I am trying to predict as an input into my table (for example the preceding value from that one that I am trying to predict, n-1 to predict n) or features that are derived from that column? I hope I have made this clear. Like I said just starting out on my ML journey and this is one thing that is causing me a few initial headaches. Any help greatly appreciated. Bonus points if anyone can provide a reference too. Cheers

$\endgroup$

1 Answer 1

0
$\begingroup$

The problem that you are describing is related to data leakage, which, roughly speaking, is the problem when injecting information about the target variable in your training data and such information will be missing when predicting new, unseen before instances with your model.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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