Girl using her phone on a translation

Machine learning for smartphone based monitoring and treatment of unipolar and bipolar disorders

Depression and Bipolar Disorder are common mental diseases and impose a high burden on society. Analysis of data collected via smartphones can help improve treatment by providing new disease insights and enabling early intervention by prediction of disease outcomes.


Smartphones provide a unique platform for monitoring and treatment of depression and mania. By replacing paper-based self-assessments of traditional treatment methods with a smartphone-based system, users can easily enter and monitor their own data and share it with clinicians who can intervene if something looks out of the ordinary [1]. Additionally, modern smartphones provide the means for collecting rich sensor data that can be analysed to provide new disease insights.

Depression and mania have long been known to impact human behaviour, such as social and physical activity. These behaviours can be passively monitored by extracting behavioural features from sensor data such as phone usage and location. Previous work in this area has shown correlations between passively monitored behaviours and mood scores [2].

Project Objectives

The project aims to predict recurring episodes of depression and mania by analysing behavioural smartphone data collected from patients suffering from unipolar depression and bipolar disorder as part of an ongoing randomised clinical trial [1]. By applying machine learning techniques, behavioural features extracted from smartphone data are related to self-reported and clinically assessed disease measures. The resulting models are then used to make predictions on new data, providing users and clinicians with new insights. Accurate prediction will enable clinicians to intervene early, improving the treatment for the individual and directing resources where they are most needed.

Important challenges researched in the project include extracting and identifying important behavioural features from smartphone data connected to depression an mania, and how to best utilise data from a population while providing accurate individual predictions from the first day of using the system, also known as the cold start problem. Using passive monitoring and prediction of disease outcomes based on sensor data alone has great perspectives and provides an attractive alternative to self-assessment because individuals do not have to constantly be reminded about their disease when things are going well. It can also mitigate issues of self-assessment adherence which are inevitable when monitoring individuals for extended periods of time.

  1. Faurholt-Jepsen, Maria, Mads Frost, Klaus Martiny, Nanna Tuxen, Nicole Rosenberg, Jonas Busk, Ole Winther, Jakob Eyvind Bardram, and Lars Vedel Kessing. 2017. “Reducing the Rate and Duration of Re-ADMISsions among Patients with Unipolar Disorder and Bipolar Disorder Using Smartphone-Based Monitoring and Treatment - the RADMIS Trials: Study Protocol for Two Randomized Controlled Trials.” Trials 18 (1): 277.
  2. Rohani, Darius A., Maria Faurholt-Jepsen, Lars Vedel Kessing, and Jakob E. Bardram. 2018. “Correlations Between Objective Behavioral Features Collected From Mobile and Wearable Devices and Depressive Mood Symptoms in Patients With Affective Disorders: Systematic Review.” JMIR mHealth and uHealth 6 (8): e165.



Jonas Busk
DTU Health Tech


Ole Winther
DTU Compute
+4545 25 38 95


Jakob Eyvind Bardram
Head of Sections, Professor
DTU Health Tech
+4545 25 53 11