0
$\begingroup$

Pardon the 101 level question. If I need to read up on some basics, please point me to few good resources.

Let's say we want to predict the Blood Sugar Level at the next 60 minute instant given a set of values for a few "features". Data is collected by commissioning a study by recruiting, say 10 people at a time and collecting their water intake (W), calorie intake (CL), carb intake (grams) (CB), exercise (E) minutes and blood sugar value (BSL). Each of the 10 patients' data is collected at hourly intervals for 15 hours daily for 3-5 days. Say the study is conducted over 3 years.

So essentially, we have between 45 and 75 measurements per study participant for 3 years.

Questions:

  1. What kind of model would I build that can predict a BSL at t+60 minutes given the values of W, CL, CB, E and BSL at t?

  2. How do I use the data to train? I dont think I can simple concatenate all the data and "build" a 3-year data set to train the model. Each study participant's data is available only for 3-5 days.

If it were a stock-value-prediction problem, I would gather years of data for a few features of the said stock as a continuous stream and use that to train a model.

Any pointers highly appreciated.

$\endgroup$
1
$\begingroup$

The time series structure should only be interesting if the single data points are NOT independent from each other and have NO trend.

For this you should be able to just consolidate the data. I´m not a doctor but after what you said you just want to use the data from period t to predict t+1. As long as there is no time trend in BSL (that would be scary) and you don´t use data from e.g. t-1, t-2 etc. you could use simple crossvalidating to train your model.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.