-1
$\begingroup$

When working with prediction problems, is there a need to consider the change in time or not?

For example, when trying to predict the price of a house, we just have the current features and the current price. The classification problem is to predict if the house price will be above a threshold or not. The regression problem is to predict the exact price.

Is there a need to have datasets with some time interval between them (for example X days or Y years) or we can predict with a dataset that was taken at a certain time? And in this case, is there a difference between classification and regression?

$\endgroup$
  • $\begingroup$ It completely depends problem at hand.. Read here datascience.stackexchange.com/q/29006/35644 $\endgroup$ – Aditya Mar 16 '18 at 0:15
  • 1
    $\begingroup$ @Aditya I can't see a relation between feature section and the above problem $\endgroup$ – Thomas Lee Mar 16 '18 at 0:44
  • $\begingroup$ I feel you opening question line says is there a need to consider time change... That comes under features. And yes you should consider them as it's natural formprices tongi up with time $\endgroup$ – Aditya Mar 16 '18 at 2:27
  • $\begingroup$ I agree that it completely depends on the problem, but I also agree that I don't see how that link is related to the question. This question is about model evaluation (downstream process), not feature engineering (upstream process). $\endgroup$ – David Marx Mar 16 '18 at 9:16
0
$\begingroup$

Is there a need to have datasets with some time interval between them (for example X days or Y years) or we can predict with a dataset that was taken at a certain time?

If you have some data that displays pattern of changing prices (or your target) over time, then sure, you have to consider the time change. This data will then become a time-series data, which has to be processed differently than normal data.

And in this case, is there a difference between classification and regression?

The problem of predicting prices at first-hand is regression, because the prices take continuous values. But you can convert it to classes according to your threshold. So in this case, (as far as I understand), there is no difference between regression and classification, as you are just putting numbers (prices) in different classes.

Hope it helps.

Please comment if something is not clear.

| improve this answer | |
$\endgroup$
  • $\begingroup$ - How to predict a house price with just one target (not time series)? For example, the only given price is in 2018 $\endgroup$ – Thomas Lee Mar 16 '18 at 12:13
  • $\begingroup$ Is seems to me that in the case of binary classification, we just need one dataset. $\endgroup$ – Thomas Lee Mar 16 '18 at 12:14
  • $\begingroup$ If you don't have data belonging to different time, then its not time-dependent. Then you should go with simple linear/polynomial regression method. $\endgroup$ – Ankit Seth Mar 16 '18 at 13:34

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.