Probably a form of wrist watch will do
the trick as far as sensors go,
detecting interaction with objects
and the environment.
Eating, smoking, drinking water,
these are predictable patterns by physical movments
limited to the composition analysis.
But that too, if taken photos of,
can be predicted of the calorie intake,
nutrition factors, behavioral patterns
by frequency (smoking),
and the rate at which the water
is consumed by the holding position toward
the mouth, (cup, bottle, or by water fountain)
whichever permutations.
point is, gather the data and list it
for observation to better provide
solutions to issues by data, or suggest them
by leads/matching algorithm.