Insights from Machine-Learned Diet Success Prediction

It isn’t often there is a huge public dataset of the quantified self data for a large group of people.

I ran across this public dataset of MyFitnessPal data for 587K days of food diary records:

MyFitnessPal Food Diary Dataset

Their paper uses this data and machine learning to predict diet success:

Insights from Machine-Learned Diet Success Prediction

I got to the paper and data via this tweet from Ted Naiman, MD and author of the book “The P:E Diet”, which relates protein intake to energy intake to explain fat gain.

How to Observe the P:E Diet in Real Life

Using basic macronutrient correlations, he shows that increasing protein content in the diet is several times more effective at calorie reduction than changing any other macronutrient.

1 Like

“Table 6 summarizes the category-based classification model. Overall, categories related to
fruit, poultry, and baked foods are indicative of staying below one’s calorie goals, whereas
wheat, pork, and fried foods point towards going over.
It is interesting to see that desserts in general (denoted by the main category “dessert”)
are associated with logging too many calories, but caramels (denoted by the specific entity
“dessert:confectionery:caramel”) are associated with logging less categories than one’s goal.
However, the average usage of caramels corresponds to only 130kcal, compared to 173kcal
for any logging entry under “dessert” and 195kcal for generic cakes (“dessert:cake”). The
appearance of donuts (“snack:snack:donut”), with an average of 180 kcal, in the under-goal
class is also unexpected” - Weber, Ingmar & Achananuparp, Palakorn. (2016). Insights from Machine-Learned Diet Success Prediction. Pacific Symposium on Biocomputing. Pacific Symposium on Biocomputing. 21.

1 Like