A major challenge with learning models is the quality of input data from consumer grade sensors. For instance, HRV data from Garmin watches, top popular choice among athletes, has been reported to have inconsistencies, as discussed in user forums [1]. If the data is noisy or biased at the source, even the most sophisticated models may struggle to produce reliable insights. Addressing sensor level inaccuracies might be just as critical as improving the learning model.
[1] https://www.google.com/search?q=hrv+site%3Areddit.com%2Fr%2F...