Digital Phenotyping-based Depression Detection in the Presence of Comorbidity: An Uncertainty Reasoning Approach

Depression is a growing health and societal problem that has become increasingly prevalent and burdensome. The detection or diagnosis of depression has been very challenging, especially for patients with other comorbidities. Digital phenotyping has emerged as a promising tool for automatic depression detection from user behavior data collected by sensors. However, existing digital phenotyping-based detection of depression has not considered the diagnostic uncertainty caused by similar symptoms shared between depression and other comorbidities, which may negatively affect detection accuracy. We propose a novel deep learning model that processes and fuses data from multiple sensors and addresses the diagnostic uncertainty based on evidence theory. We evaluate the proposed model against state-of-the-art models using sensor data. Our work makes significant contributions to design science research by proposing new artificial intelligence (AI)-based artifacts to deal with uncertainty and to mental health research by improving the accuracy of depression detection in the presence of comorbidity.