I'm trying to estimate daily household occupancy using hourly energy data.

My end-goal is to recommend a method that small utilities can use to estimate daily occupancy. So, I can't recommend ML to do the task. It needs to be something simple, rules-based, with maybe some calculations or basic regressions.

I have a training data set (whether a home is occupied on a given day). Unfortunately, it seems like a lot of energy use might not be associated with occupancy (HVAC, refrigerators, vampiric load, etc.), so finding a signal for occupancy is not straightforward.

I know that the research question is an ideal application of machine learning. My issue is that none of the methods I've looked into so far seem to give an output that easily translates into a simpler set of rules or calculations.

Is there an established method for using ML to inform a simpler set of classification rules? Any ideas of what I can look into?