I am trying to wrap my head around a customer churn problem with multivariate time series as the input data. The problem I have at the moment is, that I can not quite figure out how to pre-process and shape the data to feed it into an LSTM. I know, that the input vectors have to be 3-dimensional, but in this case I simply don't know how to tackle this specific problem.
The data contains multiple features about the users' interactions (the granularity is daily) and time-stamps about when the user subscribed to the service and when the subscription ends.
What I did so far was to mark each row (iddentified by a user-id) with 1 or 0 stating if the user churns or not after the subscription ends. This is simply done by checking if a user re-subscribed after the current subscription ended.
I think my problem lies at understanding how I have to process the data in order to get feature vectors which I can input into an LSTM in the end to get a class-prediction for each user.
Any help is very much appreciated!