Clustering of Functional Data to Discover Patterns of Behavioral Trajectories Using Smartphone Data

• Activity Number: 567 - Digital Phenotyping – What Can Wearables and Smartphones Tell Us About Our Mental Health?
• Type: Topic Contributed
• Date/Time: Wednesday, July 31, 2019 : 2:00 PM to 3:50 PM
• Sponsor: Mental Health Statistics Section
• Author(s): Samprit Banerjee* and Jihui Lee
• Companies: Weill Medical College, Cornell University and Weill Cornell Medicine
• Keywords: digital phenotyping; functional data; clustering

Abstract

Passive collection of data on an individual’s activity via smartphones or wearables can provide a unique insight into the behavior of patients with mood disorders (e.g. depression). We propose a functional data framework to analyze passively collected data on activity and socialization. An improvement in the depressive mood of depressed patients can be signaled from their improvement in measures of activities and socialization. To understand patterns of change one needs to cluster the behavioral trajectories of activity and socialization into patterns of change over time i.e. clustering of functional data. Traditional methods of clustering of functional data cannot be interpreted easily by a clinician. We propose a new method to cluster functional data based on the derivatives of their functional forms. We discretize functional derivatives into states of increase, decrease or staying the same in order to avoid distinguishing patients on their baseline levels. We will present simulation studies and application on real data to illustrate the performance of our functional clustering methods.