1
$\begingroup$

I have the Data of 10,000 users Time Session in a website/App, The Login time, logout time, the person activity,

The Data is available for 60 days ( per user )

Using this 60 days data for 10k users, can I predict the active time of the respective user on 61st day? if yes what is the best approach and please suggest which type of problems I can refer to solve this

Thank you

$\endgroup$
3
  • $\begingroup$ What do you want to do with that information? Knowing how the prediction will be used may help people give you good advice. $\endgroup$ Apr 15 '20 at 15:45
  • $\begingroup$ What you have is a regression problem. That gives you two options depending on the data => time-series model or an ML model. I'd go with an ML model $\endgroup$ Apr 15 '20 at 19:52
  • $\begingroup$ are there 10K distinct user ? (I assume no) - how many users (different user_id) do you have ? $\endgroup$
    – user702846
    May 28 at 11:47
0
$\begingroup$

Convert your training data to the following format -

a = login_time
b = logout_time
c = person_activity
train_data_X = [
               [[a, b, c], [a, b, c], ......, [a, b, c]]
               [[a, b, c], [a, b, c], ......, [a, b, c]]
               .
               .
               .
               [[a, b, c], [a, b, c], ......, [a, b, c]]
             ]

train_data_Y = [
                 [[time_day_1], [time_day_2], ...... [time_day_60]]
                  .
                  .
                  .
                 [[time_day_1], [time_day_2], ...... [time_day_60]]
                ]

Explaination -

You Input(training_data_X) data has shape 10000 * 60 * 3 because you have 10K users, data for 60 days and each data point depends on 3 variables namely a, b and c. Your Target(training_data_Y) data has shape 10000 * 60 * 1 because you have just one number to output i.e active time.

Now, this is a Regression Problem because you are trying to predict a continuous value i.e active time.

You can model this type of problem with a number of time-series models. But since you got a multivariate input, i.e [a, b, c](refer training_data_X). You would be benefitted a lot using a Seq2Seq model to generate your time-series. If just want to generate 61st time-step you could use it just for that as well.

Specific details and implementation

$\endgroup$
7
  • $\begingroup$ you don't have user_id - do you ? then how does the model learn about each user_id activities ? $\endgroup$
    – user702846
    May 28 at 11:45
  • $\begingroup$ The first dimention i.e the 10000 rows or in this case matrices are your users. Assign any id to them. $\endgroup$ May 30 at 22:01
  • $\begingroup$ 1) I don't see any user here -> a = login_time b = logout_time c = person_activity - yes, these are user information but do not specify which user - even if they are user_id, how does that gonna look after all those numerical transformation ? - 2) the link you have added at the end, is not repeated measurement - every household is in one row - in this question one user is not necessarily one row $\endgroup$
    – user702846
    May 31 at 7:34
  • $\begingroup$ 1st row is first user. 2nd row is second user. Like sentences words and characters. $\endgroup$ Jun 1 at 8:17
  • $\begingroup$ Only difference between the link I posted and the problem here is I additionally suggested dividing it day wise adding one more dimension to the "user" row $\endgroup$ Jun 1 at 8:18

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.