Health Data to predict Glucose?

Hey everyone! I’m trying to learn machine learning and I thought a fun project would be predicting glucose. Has anyone ever tried using Apple Watch / Fitbit / Garmin data to *detect or predict* glucose changes? Not as a replacement for a CGM, but more like a rough early warning / risk signal using stuff like HR, HRV, temp, sleep, activity, stress, etc. I originally asked in diabetes subreddits, but most people there have CGMs (and even CGMs can be noisy/inaccurate). I’m more thinking: could this be researched as a way to screen for prediabetes or flag “something’s off” trends, even if it’s super imperfect. “A here’s why this won’t work” also welcome. Thanks!