The article you cite states that accuracy drops to 60% if the enrollment and testing data were collected at different sessions. I imagine the effects of coffee or walking on heart rate would make that even worse.