I have data on customers' usage of various product features over time. Each month, a customer can choose to use a feature or not.
I want to create a live system that produces the probability of a user continuing to pay for a feature at the end of the year. To do this, I want to consider whether the user used the feature each month (constantly updated) and their demographics (which change slowly or not at all). This means that I want to update the probability of payment at the end of the year each month. For example, after a month with no usage, the yearly probability will decrease; after three months of consecutive usage, it will be very high.
I have difficulty deciding which model to use and how to structure the data (cross-sectional, time-series, survival, panel). I think this is because: a. The outcome level is at the yearly level, but the usage data (and needed probabilities) are at the monthly level. b. The usage status is not changing but is added (i.e., a binary value for that month's usage could not be filled before).
Do you have any ideas on how to approach this problem?
