Can a machine learning model be trained on Call Detail Record(CDR) Data to predict user's daily locations?


I have a CDR data for two months and my goal is to extract daily or frequent locations(cell towers) of the user along with the departure and arrival time on those locations. The spatial resolution of CDR data is coarse since it is only generated whenever user makes any call or sends a sms and also it does not have the actual coordinates but the coordinates of the cell towers. I believe that it is a clustering problem but I want to ask that can I train a model that is intelligent enough to filter out noise i.e the cell towers that are rarely contacted or locations that are not frequently visited, and identify the locations and estimate departure and arrival time. Also is there any model that learns from the spatio-temporal historical data that can be used for predicting such locations?


Posted 2019-12-14T09:43:59.123

Reputation: 101



Count all the cell towers the user is in. (Use durations as weights).

Keep the maximum (or largest k values) only.

Doesn't this trivial counting already solve your problem? Why not?

Has QUIT--Anony-Mousse

Posted 2019-12-14T09:43:59.123

Reputation: 7 331