6
$\begingroup$

I'm trying to measure galvanic skin response in real-time with a wrist-wearable device, similar to what is being done in the paper "Using Noninvasive Wearable Computers to Recognize Human Emotions from Physiological Signals" which is used the, now discontinued, BodyMedia SenseWear. What solutions exist for this problem? From my preliminary research, I have found:

Are there other solutions that I've missed? How does one compare the accuracy of these various solutions?

$\endgroup$
  • 2
    $\begingroup$ Related: Useful tools for analysis of EDA/GSR/GSL. cogsci.stackexchange.com/questions/16017/… $\endgroup$ – Robin Kramer Nov 10 '16 at 8:28
  • 2
    $\begingroup$ I had Jawbone 3, I was unable to get GSR data. (choosed only for the GSR) (I'm suspicious about a 'feature bluff'.) $\endgroup$ – Bálint Kőszegi Nov 12 '16 at 19:16
  • 1
    $\begingroup$ iMotions does not make their own GSR devices. They integrate with the Shimmer (worthwhile looking into since you did not list it) and Empatica devices (and allow integrating others). Should be clear now on the page you linked—perhaps it was not at the time of writing. P.s. I work at iMotions. ;p $\endgroup$ – Steven Jeuris Aug 15 '17 at 8:46
4
$\begingroup$

iMotions has a pocket guide on GSR targeting a layman audience (believe the PDF you can request might be more up-to-date). One key thing to point out is (emphasis added):

Our body has about three million sweat glands. The density of sweat glands varies markedly across the body, being highest on the forehead and cheeks, the palms and fingers as well as on the sole of the feet.

For wrist-worn devices this means ideally you place the electrodes on the palms or fingers. Measurements on the wrist are much less sensitive, as shown in Seanny123's answer in case of the Empatica E4. Of course, not all experiments allow to place electrodes there, in which case electrodes on the wrist might still be worthwhile exploring.

Wearable devices supporting electrodes on the palms/fingers:

Watch-like devices, with electrodes placed on the wrist:

Other wearables:

Disclosure: I work for iMotions. Feel free to edit this list and include any other wearable devices you know of.

$\endgroup$
  • $\begingroup$ Great list! Do you know about information on accuracy of palms/fingers solutions compared to laboratory quality equipment? It would also be helpful to add approximate costs for each device. $\endgroup$ – user2503795 Dec 6 '17 at 18:23
  • $\begingroup$ @user2503795 I believe this would make a good new question! $\endgroup$ – Steven Jeuris Dec 6 '17 at 18:33
  • $\begingroup$ Here you go $\endgroup$ – user2503795 Dec 6 '17 at 19:13
4
$\begingroup$

A system that I am currently testing is the MoviSens EdaMove. The system consists of a small device that can be attached to the arm, and two electrodes to measures EDA at the palms of your hand. The EdaMove uses the exosomatic method, by applying a 0.5V direct current, and samples with 32 Hz. An accelerometer is also included in the sensor, so that you can correct the EDA recordings for movement.

So far, the recordings are pretty good, but I haven't done any formal testing yet. Usability-wise, it may not be the best solution, since it is somewhat more intrusive than a wristband/watch.

Update: This is the fun part. Skin conductance is a measure of skin sweatiness. Unfortunately, the stickers that attach the sensors to the skin (similar to how ECG/EEG-reference sensors are attached) let loose when the hands start to sweat. This is even more so with many movements. For tasks that require much manual work or fine hand-movements, the EdaMove may not be the best sensor.


From the E4, I know that performance is not perfect, but it is okay. Benefit is that the E4 also includes an PPG sensor and accelerometer. Moreover, the E4 allows you to monitor the recordings in real time with their Android/iPhone app.

There are two major issues though with the E4. 1: It needs a good connection with the skin. If not, you could rather easily have missing data (especially so for the HRV sensor). The sensor should thus fit rather tightly. 2: Empatica applies some filtering of the data before you can extract it. It is not clear, however, to what extent it is being filtered. NB. The website states that raw-data is provided. Only the interbeat intervals (IBI), an HRV metric appears to be filtered.

$\endgroup$
  • $\begingroup$ How much does the EdaMove cost? $\endgroup$ – Afflatus Jun 1 '18 at 18:38
3
$\begingroup$

Accuracy Comparisons

In Simon Ollander's thesis "Wearable Sensor Data Fusion for Human Stress Estimation" the Empatica E4 sensor is compared with laboratory quality equipment. Specifically, in the case of GSR, these were finger-tip electrodes as shown below:

electrode placement

Two tasks were compared, D2 and SCEPT. The SCEPT task consisted of the subject putting their hand in 0 freezing water (stress) vs. warm water (control), while the D2 required the subject to find all d:s with two lines among other symbols under time pressure (stress) vs. no time pressure (control). As you can see from the plots below, the signals collected from the Empatica E4 sensor differ from the lab equipment:

d2-task

scept-task

To determine whether these variations were significant a stress classifier was trained on the lab-quality equipment and validated on the E4 data.

results-table

Not sure what happened with that Naive Baye's classifier...

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.