Motion and ambient temperature contaminate the signal

What it is

Heart rate on a watch and electrodermal activity at the palm or wrist are easily rewritten by walking, typing, and room temperature. Those changes travel through motion artifact and thermoregulatory sweating, not emotion or cognitive load. Contamination can outgrow the target signal. Without independent motion and temperature channels, a classifier reads a flight of stairs as a stress peak.

Why it happens

Optical heart rate (photoplethysmography, PPG) uses light absorption by blood volume under the skin. If the sensor slides, or muscle contraction changes the optical path, the waveform gains components at gait or gesture frequencies, and peak detectors treat them as beats. EDA at palm and sole is relatively sensitive to psychological stimuli; on the dorsal wrist and forearm it is more thermoregulatory. Ambient heat and metabolic heat raise tonic skin conductance and look like sustained arousal. Limb swing also changes electrode contact area, producing steps that are hard to tell from a skin-conductance response. Heart rate itself has an exercise physiology: cardiac output rises with metabolism. That is autonomic regulation, not interface stress.

Studying it

Place the same sensors in seated elicitation, walking, typing, and a warmed room or seasonal contrast. Factors: walking speed, ambient temperature, site (palm versus dorsal wrist). Outcomes: heart-rate error against ECG or a chest strap, false EDA responses in periods known to have no psychological event, and classification change with versus without accelerometer regression. An “emotion model” collected in a still, constrained lab will collapse on a walk. Reports should keep acceleration and temperature as covariates rather than deleting “moved” trials after the fact.

Where it stops holding

A chest-strap ECG resists hand motion better than wrist PPG, and still cares about torso motion and muscle artifact. On a desk task with stable air-conditioning and almost still hands, thermal and motion contamination can be small, and laboratory conclusions hold relatively well. Palmar EDA tracks psychological stimuli better than the dorsal wrist, and is also easier to ruin with a handshake, sweat, and contact-area change. Swimming, showers, and winter outdoors are extreme contamination, not a slight accuracy drop. Drug-induced sweating and fever impersonate arousal as well.

Applying it

• Before heart rate or EDA may drive an interface decision, check on-device acceleration and temperature; freeze inference when motion or thermal slope exceeds a threshold, rather than emitting an emotion label.
• State measurement site in the capability copy: wrist optical heart rate is unreliable while typing or swinging the arm.
• Prefer palmar laboratory electrodes when cleaner EDA is required; treat a consumer watch as a coarse arousal trend, interpreted only together with activity type.
• Verify by running the same psychological task while walking, typing, and entering/leaving air-conditioning, against a seated baseline, and count how many “stress events” appear only with motion or temperature change.

Related

• Same group: C9.03.1 Physiological measures reflect arousal rather than specific emotions · C9.03.3 Inferring state from physiology and changing the UI requires informing the user
• Adjacent: C9.09 Wearable Fit and Signal Quality · C9.04 Respiration and Posture Sensing
• Search: PPG motion artifact · thermoregulatory sweating · tonic EDA