I have the following data at hand:

  • data about internet usage, per hour, per user, per part of the day (morning, afternoon, evening);
  • the category of websites visited and their duration;
  • demographic information about the users;
  • (basically all surfing information that you can think of is available).

I have 1000 users with their corresponding data for 24 months. Approximately a total of 24 x 1000 data points (per user per month).


I am trying to predict the trend of web surfing for a group (not an individual), for about 6 months ahead in time, based on several demographic and historical features. I could just plot the surfing duration for porn websites over time for teenagers and call it a trend, but I also need to take into account the impact of other features.

I earlier thought of this as a regression problem, but I realised that I could only predict for a given user at a particular point in time, based on historical data. This wouldn't give me a surfing trend for categories or age groups.


  • How can I find such trends for a category (e.g., adults and porn) based on other demographic information.

  • How can I see the interactions between the variables? What is the approach or the models I should look at?

  • $\begingroup$ I have trouble identifying the issue here. You can predict individual users' behaviour and sum the predictions to obtain a prediction of the group's behaviour. You can sum the individual users in the beginning and model the behaviour of the whole group and forecast it as such. You can include as many features as you like in your models, be it vanilla regression, regression with ARMA errors, a neural network, etc. $\endgroup$ – Richard Hardy Jul 23 '16 at 12:23
  • $\begingroup$ Thank you Richard, I have been able to extract very useful information from the last few of your answers on my questions. I'm sorry i haven't been able to give you feedback on each of them. To accommodate multivariate time series, I tried VAR and ARMAX as you described in one of your responses but the model significance was bad. So i ended up having a seasonal uni variate ARIMA model. Thank you once again. $\endgroup$ – Arslán Aug 2 '16 at 14:48

Your Answer

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

Browse other questions tagged or ask your own question.