How to deal with choice of population of study?

[KaliHD]

Hi everyone,

I am glad to find a forum like that.

I am currently looking for doing classification or clustering of my big data set on web application.
I would like to do cluster on websites visitors who are identified by their cookie ... which they can drop whenever they want.

The problem I am to deal with is the life time cookie! To get an idea I have about 3 millions "unique" (if they drop their cookie a customer can be identified as new another time) visitor and 85% where the information related to the "unique" visitor is just for one visit and 15% where the information can be collected over more than 1 day (typically about 12 days).

I don't know if it is a good idea to take into account in my clustering (or classification) the 85% of visitors and to do my model on "just" the other 15% which seems more relevant?
Or is this dummy? Because yes they can drop the cookie and yes if I take into account the 85% maybe I will see the same visitor 20 times but with 20 different "unique" key because he has drop the cookie each time.
But an another side the visitor can also come only one time and never come back again?

Have you maybe an idea also on which articles I can find information and ideas about that?

All the best,

KaliHD

[editor]

It is a tricky question, but I would start by building 2 separate models - one for multi-time visitors and one for 1-time visitors.
If 1-time visitors visit similar types of pages, then perhaps you can apply multi-visit model to one-time visitors