I recently counter an assignment which is to predict whether users will click ads shown on some websites in the next week based on users' log history. The log contains users' id, os and browser type, time, website id, ad id, click or not, etc...

The problem is that this isn't a classification problem which I'm familiar with, because I don't know users' behaviour data in the next week(I mean there is no test dataset). So I'm confused and don't know where to start.

Should I simulate users' behaviour data(like when and uses what os and what browser to browse the website and so on...) first and then transform it into a classification problem? But how to simulate?

Any idea is welcome, thanks in advance.


Simulation is probably not a good idea because you need to generate a function $f$ that will predict the presence of clicks. In the end, you performance will just measure how well you can approximate this particular $f$ that you simulated (and already know).

You can either report the scores obtained on cross-validation (the method is up to you) or you can split your train set into an actual train set, and a test set. The labels of the test set are just used to evaluate the performance of the model.


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.