New Study Predict How Much A Person Will Eat From Smartphone Data

Increasingly smartphones are being used be people to log various aspects of their life. Recently my group published a paper in which we analyzed data from individuals’ smartphones, to see whether we could model their dietary patterns. Rather than compare across people, we focused on the patterns that exist for specific individuals. We examined whether the portion size for each person’s meal tended to followed a routine based on time of day, whether the portion size was related to how much physical activity they engaged in before the meal, or whether the their “food environment” — the quantity, diversity, and types of food establishments around them — influenced the portion size. We also looked at the person’s mood to see it influenced their meals. All of these data were collected from a smartphone app we developed (voice-annotated video recording of meals, accelerometry for physical activity, GPS for food environments, and occasional prompts to ask about mood).

What we found was fascinating. In some cases, certain individuals exhibited fairly routine eating patterns, while for other individuals, exercise before the meal was an important factor in determining their portion sizes. However, we were surprised to find that across all of the subjects, the food environment tended to be an important factor for explaining portion sizes.

This study was based on data from a fairly small group of test subjects who were only monitored over a short period of time, but we now have hundreds of gigabytes of similar data for which we will be doing the same sort of analyses.

The work provides an early glimpse of our ability to integrate various different sorts of sensor, self-report data, contextual data to model specific individuals behavioral patterns.  Moving forward, we are interested in evaluating how well these models work for prediction, and eventual use for tailoring behavioral interventions to specific individuals to help them live healthier lives.

This study was recently published in the journal PLoS ONE: http://dx.doi.org/10.1371/journal.pone.0153085

 

Comments are closed.