I have a pandas dataframe that represents a time series. My time series is segmented over the phase type that the robot is performing (i.e. I have a column with the phase type per timestamp and the phases are known). Some easy examples: if the machine is cleaning something the phase is "cleaning", if It is moving the phase is "moving".
I have no domain knowledge about the phase in which I am and the constraints in the value that each signal must respect. I am searching for rules with a certain confidence.
I would like to say: I am in this phase, then from the data I have seen in the past, I know that for sure signal A will be less than signal B with 90% confidence. Or again, signal C should not be negative according to what I have seen in the past with 70% confidence. I want to extract historical simple rules that I would like to validate at the end of the process with a domain expert.
Is there any library or method that can handle this type of problem? I didn't find much online. It seems like association rule mining, but I am working on time series and I am looking at time periods spanning the whole phase, so a very long time period.
Otherwise, if I should pass from other methods as correlation/cross-correlation analysis between time series, can you point me out the best analysis I should use? And also explain to me how should I exploit my analysis result for transforming them into more simple rules like the one I have described before.
