Visual analysis of relationships between behavioral and physiological sensor data

More and more people are collecting, organizing, and interpreting health data through off-the-shelf products such as Apple's Health Kit. People use these systems to collect steps taken, calories ingested, etc. These ecosystems also support the collection of physiological data. For example, users collect heart rate data during exercise, and even electro dermal activity data to help detect the onset of seizures. Analyzing physiological data, however, and relating it to specific behaviors or events, is challenging. In this paper, we present an 11-week, multi-session, participatory design case study, identify challenges in analyzing physiological and behavior data, and present BEDA, an analytics tool we developed to mitigate the challenges. The two primary data analysis challenges include: (1) interfacing multiple software programs required for capturing and analyzing the different data sources, and (2) extending the limited data analysis functionality within and across these software programs to support a wide range of physiological data analyses. BEDA resolves the fragmented analysis pipeline by integrating closely-related analysis tasks into a common interface. It also addresses the extensibility problem by integrating scripts that apply any custom or publicly-available function written in MATLAB or R. These scripts extend basic analytic capability, provide the analytic bridge between physiological and behavior data, and incorporate machine learning algorithms to highlight behaviors associated with physiological data. BEDA's capabilities mitigated the challenges of signal analysis and fragmented data sources, and motivated behavioral scientists to combine physiological measures with behavioral analysis. Although we developed this tool for a domain-specific case study, the use of the tool can be generalized to analyze any time-based data source or sources.