We suspect that some of the data in the REAL-PD dataset is corrupt. There seem to be closely spaced samples which take very different values (as if two different data streams are sampled) leading to a "jigsaw" effect. For example 5b4c7c81-659d-40ea-a1fd-59622074fd10 in the smartwatch_accelerometer directory: 1.28 9.782708 -0.454898 2.796427 1.282 -8.465898 -4.683059 1.733402 1.302 9.782708 -0.454898 2.796427 1.303 -8.374918 -5.214571 2.001553 1.322 -8.259995 -6.196194 2.303222 1.342 -8.149862 -7.201759 2.461239 1.344 9.773131 -0.397437 2.825158 1.361 9.68694 -0.502782 2.791639 1.362 -7.805098 -7.61835 2.748544 This exists also for the smartwatch_gyroscope data.

Created by Izhar Bar-Gad bargad
@trunghieuhoang - All REAL-PD Subjects wore Motorola Moto 360 watches.
Hi @sieberts , Do we have any metadata for the correlation between the deviceID and the actual smartwatch model used by the subject? I mean is it possible to have the brand name and the model of smartwatch devices used in the REAL-PD dataset? Thanks
@jwolson The device ID indicates which device the measurement came from. The subjects were instructed to wear one watch at a time, but you can see that in some instances they were wearing two at once, which resulted in the original observation of seesawing in the data. We don't have any annotation as to which limb each device was worn on.
@sieberts Nevermind this comment
@bargad- We are trying to go back to the data curators to determine the extent of the issue. We'll keep you posted as we know more.
@atpage - Acceleration in the REAL-PD files are measured in m/sec^2^, so ~10 m/sec^2^ would be the acceleration of gravity. We'll clarify the units on the data description.
In addition, 10 g's of acceleration for 20 minutes straight doesn't make sense. That's what I'm seeing in c15365b5-fa46-45c4-94a1-802ac74bef82.csv (the first one I opened). What are the actual units? m/s^2?
It is a sample from the smartwatch accelerometer and not the smartphone...
@bargad- We are looking into this issue.
Is this sample from real-pd/training_data/smartphone_accelerometer/5b4c7c81-659d-40ea-a1fd-59622074fd10.csv ? I don't see these particular measurements for 1.28 <= t <= 1.362.

Corrupt smartwatch measurements in REAL-PD page is loading…