We presented the design and evaluation of BES - a sensorbased computational model that provides daily automated
sleep duration monitoring using commodity smartphones. Unlike existing consumer products (e.g., wearables, smartphone
apps) and research prototypes, BES requires zero behavior
changes from the user; individuals do not have to change the
way they sleep, signal to the phone when they wake up or
attach specialized sensors to their body. Rather, BES relies
on the collective predictive power of a series of soft sensorbased hints that relate user behavior to sleep duration (e.g.,
prolonged silence or the smartphone remaining unused and
completely still). From a preliminary one-week 8-person study
we find BES is able to exploit these hints to estimate sleep
duration with a surprising degree of accuracy (± 42 minutes)
and provides sleep duration estimates close to commercial
alternatives requiring wearable sensors. Our results contribute
towards the on-going development of mHealth technology able
to track key behavioral dimensions that impact overall health
and wellbeing; but yet must also remain unobtrusive to users
and suitable for daily long-term use.