> When 42 of the participants came back to the laboratory weeks, months and even two years later, to take part in another 24-hour measurement, the trained algor

> When 42 of the participants came back to the laboratory weeks, months and even two years later, to take part in another 24-hour measurement, the trained algorithm could identify them from their breath patterns. Data from periods when the participants were awake gave more accurate results than did those from sleeping periods, but when the researchers used a 100-parameter characterization of a full data set instead of one using 24 parameters, they could pick individuals out with 96.8% accuracy. The correctly identified .968x42=40.696 of the participants. Also any standard-ish physical activity that comes with instructions usually includes breathing in those instructions. So I would expect results to vary substantially depending on where they found the study participants.