A developed system able to classify the states of patients suffering from bipolar disorder, a severe chronic psychiatric problem, is introduced in this paper. Commonly, the diagnosis of this disease is based on clinical evaluations through interviews and estimations of scores gathered by quantitative psychopathological ratings. Although well established and defined, these evaluations have their drawbacks as they are performed on sporadic days, while a change to a potentially dangerous state can happen in between.

Therefore, timely monitoring is relevant for classifying critical states of the disorder. Smartphones present an enabling technology for this purpose. They can utilize embedded audio and accelerometer sensors for patterns of physical activities and for social interaction tracking, and provide periodic delivery of self-assessments by the patients. The obtained data was compared with outcomes from psychiatric evaluations. Average classification accuracy of over 80 percent was achieved for all precision and recall values. This allows for giving reliable information to clinicians that could take preventive actions.

The importance of the presented paper is twofold. On the one hand, it shows medical workers a good quality example of how to organize continuous monitoring of patients with severe health conditions. On the other hand, this paper is of interest to information technology (IT) specialists as a particular construction of a purposeful system employing mobile computing.