I have a dataset of time series data that represents a process, with the total cost increasing over time. My challenge is to determine if litigation will occur, which will drastically increase costs.
My dataset includes the total cost of the process as recorded each day that the process was ongoing, and the number of days between when the process started and the process entered litigation (once it's in litigation, you can't flip the flag back).
The data set is shaped like this:
|key|day|value|in_litigation| |a |1 |20 |False | |a |2 |25 |False | |a |3 |40 |True | |a |4 |66 |True | |b |1 |4 |False | |b |2 |11 |False |
What's the best way to build a neural network to solve this? In the training set, let's say that I have a process where it went into litigation happened on day 15. If my test process is on day 10, I would think that I would want to compare my test row with what the training set looked like on or before day 10 and use that data along with the knowledge of whether the event happened to predict when it could happen in the future.