Data Processing Pipeline of Short-Term Depression Detection with Large-Scale Dataset

Abstract

Depression is a common, recurring mental disorder that causes significant impairment in people's lives. In recent years, ubiquitous computing using mobile phones can monitor behavioral patterns relevant to depressive symptoms in-the-wild. In this paper, we propose data processing pipeline of short-term depression detection using mobile sensor data. We build a group model classified by depression severity for capturing depressive mood in a short-period time to handle data quality and data imbalance problem in a large-scale dataset. We expect the group model to identify and characterize digital phenotype representing each depressive group as a middle step toward personalization.