Blood Sugar Prediction
Using explainability to reduce sensor costs
Imagine we have a device that collects data to predict blood sugar levels accurately. We could attach sensors to measure:
- Temperature
- Blood sugar
- Heart rate
- Blood pressure
After training and analyzing, we discover that we only need the blood sugar sensor to predict blood sugar levels with the help of explainability.
Obvious in this case, but in most cases, we do not know which combination of sensors predicts something reliably enough.
So we need to find out and then remove the unnecessary heart rate and blood pressure sensors. This:
- Reduces production costs
- Creates a lighter, more comfortable device
So simply makes the device more accessible to diabetes patients.