Can My Network Tell How I Feel?

Your router sees the traffic of every device in your home. The content is encrypted by default, but when you talk, how much, how often, and roughly to whom is still visible. This metadata can give clinicians insight into the behavioural criteria used to diagnose depression, from the router alone, with no app and no wearable: whether you are awake at 3am, how varied your activity is, how stable your routine looks.

Clinicians cannot act on an indicator that arrives without a reason. If the system reports signs of disturbed sleep, the chain of events behind it has to be clear. Our approach, CareNet, follows this principle with explicit rules and weights instead of a black-box model. So far, though, it has only run offline on stored captures: neither a local deployment nor the explanation chain has been demonstrated on live traffic in a real house.

This project turns the analysis into a local, explainable appliance. A MikroTik router mirrors household traffic to a mini-PC where the whole pipeline runs, so no data leaves the home. The students script daily routines that vary wake time, nocturnal activity, activity diversity, and routine stability, record the real traffic each one produces, and test whether the pipeline recovers the routine and whether every reported likelihood can be traced back to the packets behind it, in terms a clinician can follow. The only behaviour recorded is the scripted one, so there are no patients and no personal data. The goal is to make a future field study with real patients possible.

The project builds on Stephan Nef’s Master’s thesis and Tibor Haller’s Master’s thesis on explainable depression awareness at the edge.