Mobile Devices: Behavior Monitoring of Mental Health

Ubiquitous sensor rich mobile devices are massive overkill for the needs of most people most of the time. And there are certainly downsides to carrying a poorly protected surveillance device on you at all times. But the ubiquity has some potentially valuable side-effects, if they can enable unobtrusive behavior monitoring when needed (and requested).

Specifically, sensor rich devices may help watch for behavioral symptoms of mental health problems, such as depression. Sandy Pentland discussed this concept earlier (see his book Social Physics and citations), and devices have even more capabilities and universal acceptance now. People already carry the device, so all you have to do is add the right app. No special “monitor” needed.

Venet Osmani of CREATE-NET in Trento, Italy reports on recent work that uses sensor data from mobile phones to detect behavioral changes associated with the mood changes diagnosed as Bipolar Disorder.  Basically, they use machine learning methods to discover patterns that reliably correlate with the psychiatric and subjective observations (with 90% plus accuracy).

The data collected include GPS and accelerometer data, which reveals “activity level”, and also voice data when available. The latter may reveal changes in speech patterns indicating stress or depression. Osmani’s report did not indicate that social networks were used, but I could certainly imagine that social network analysis might add useful patterns. (E.g., reduced contact with friends and families during depression.)

This work is interesting, and shows both promise and limits of the approach.

First of all, the goal is really to detect subjective signs of what may be very transient problems. (As my teachers taught me, crazy people aren’t crazy all the time.) Worse, the symptoms are really exaggerations of perfectly ordinary behavior, and often have to be recognized against the normal behavior of the individual person.

Second, the data available to the mobile device is mostly indirect. It isn’t detecting actual brain activity or emotions, just the visible side-effects, if any. This is something that machine learning is pretty good at, which is why it actually works. Given enough data, we can find patterns!

Third, the ubiquitous device can monitor 24/7 and in normal life. This is extremely useful for helping people live well, yet let them be watched over for signs of trouble. It is difficult to overstate the value of the unobtrusive nature of this monitoring.

The hope is, of course, that detection and tracking can help people deal with problems, and get help if they are in trouble. If this kind of monitoring were incorporated into therapy it might be possible to notify loved ones and therapists if a person appears to be slipping into depression, say, even if they don’t recognize it. Perhaps we can design human interventions, e.g., if I get a ping from the app, I text and call my friend, just to see what’s up.

This technology will also be useful for checking on the efficacy of therapy of any type. If the therapy team can get reports in the following weeks to say “its working” or not, they can adjust the approach. Certainly, this should be used to check and adjust the use of pharmaceuticals (which are notoriously difficult to get “just right” for each individual).

Osmandi’s group is working on Bipolar Disorder, which is actually one of the easiest disorders to detect. The behavioral signs are obvious, if subjective, and the behavior is chronic and cyclical.

It will be more difficult to apply these techniques to other problems, because the signals are less clear, more tangled in the “noise” of everyday behavior, and possibly triggered by unpredictable events and situations. Also, there are many problems, such as depression or drug addiction, where detection isn’t really an issue: we know it’s there, we need ways to help.

Still, this kind of monitoring might help people, especially if it can detect things that “make it better” and help them do more of that. For example, if an app can notice that the individual is a bit happier if they walk to the store (and, by the way, stop watching TV for a little while!), or meet with a relative. In this case, the person might want to be prompted to do that again. And so on.

Finally, I would note that no amount of computer surveillance of any kind is going to really solve the need for human interaction, love, security, food, and so on. The most successful responses to psychological troubles always involve human contact and caring. A mobile app is not a substitute for caring. If phone based monitoring is deployed “instead” of therapy, social support, or family contact, then it will do little and probably make things worse.

Future work will be needed to validate these techniques in more cases and in real life, and to demonstrate that it is safe and effective. But this is a good start.


  1. Osmani, V., Smartphones in Mental Health: Detecting Depressive and Manic Episodes. Pervasive Computing, IEEE, 14 (3):10-13, 2015.

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Google+ photo

You are commenting using your Google+ account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )


Connecting to %s