I would add that once you get rid of the idea that the body scale should display directly what it measures, you can also do things like menstrual cycle controlled weight for woman.
(Another idea along those lines I once had was controlling for days of the week, as many people eat more on weekends than on weekdays. But the more parameters the model has the more likely overfitting becomes.)
Another idea along those lines I once had was controlling for days of the week, as many people eat more on weekends than on weekdays.
If you make 7 day rolling averages you already remove that effect. In any case from the weight charts that I personally looked at that effect doesn’t seem that big. Day to day changes in water content seem to produce more noise.
(Another idea along those lines I once had was controlling for days of the week, as many people eat more on weekends than on weekdays. But the more parameters the model has the more likely overfitting becomes.)
If you make 7 day rolling averages you already remove that effect. In any case from the weight charts that I personally looked at that effect doesn’t seem that big. Day to day changes in water content seem to produce more noise.