I'm working in a team that is collecting data by bicycle : We have biometric t-shirts that measure our ventilation rate. The problem is that during the last data collection, participants used masks to protect themself against air-pollution.This has significantly affected the measure. So, I have a variable Y (ventilation rate), collected for three participants with a time resolution of one minute in previous data collections. For the same participants, I have new data but "corrupted" by the effect of the mask.

Is there a way to clean the data "corrupted" by knowing previous data about participants ? I have some variables linked to the ventilation rate like the speed of the participant, the mean slope of the terrain during the minute measured, the mean acceleration etc.

I found the third part of this blog interesting : https://abidlabs.github.io/removing-noise-from-signal/ They talk about "contrastive Dataset" and "confounding signal"

If somone has an idea, you are welcome !

All the best

  • $\begingroup$ Which is the final purpose? Which are the questions you are about to answer? $\endgroup$ – rapaio Oct 25 '18 at 18:11
  • $\begingroup$ @rapaio is right. Its difficult to know what approach is best without knowing your goal/questions. But if you're just looking for ideas, you may want to looking into a mixed-model approach and assign random effect to a variable "wore mask" $\endgroup$ – JustGettinStarted Oct 25 '18 at 18:22
  • $\begingroup$ Hi guys, and thank you for the responses. So I would like to remove the effect of masks because in a second time, we are analysing the product of the ventilation and measures of air pollution to have an idea of pollutant intake by participant. But here, the mean of the ventilation is higher due to mask effect. So I would like to correct these effect to have a more precise estimation of air pollution intake. $\endgroup$ – jérémy Gelb Oct 26 '18 at 0:59

Quite naive approaches could be:

  • Calculate the mean of data with and mean without the mask. Substract the difference of the means from the data with the mask.
  • Calculate the mean of data with and without the mask. Calculate the ratio of them. Multiply the data with the mask accordingly.

If you know who was on the bike, you can try this manipulation on separate subsets of the data.

  • $\begingroup$ Considering to remove the effect of the mask as a mean by participant is the "worst option" I'm considering now. I think that the effect of the mask is correlated with the variables affecting the physical activity of the participants (like the speed, the slope and the acceleration). I would like to use all these informations to "filter" the effect of the mask in a more precise way. But thank you for the response. $\endgroup$ – jérémy Gelb Oct 26 '18 at 1:03
  • $\begingroup$ Hmm...This is difficult because pollution itself likely has an effect on the ventilation rate but that effect was diminished by the mask. $\endgroup$ – Isabella Ghement Oct 26 '18 at 3:29
  • $\begingroup$ @jérémyGelb What are all the variables that you observe? What are all the variables that you consider but cannot observe? $\endgroup$ – Karel Macek Oct 26 '18 at 5:17
  • $\begingroup$ @KarelMacek : the observation unit is a segment of one minute, and the variables are : the ventilation rate, the heart rate, the speed and the acceleration of the cyclist, the slope of the terrain, the No2 concentration in the air. We consider the effect induced by the participant as a random effect. $\endgroup$ – jérémy Gelb Oct 26 '18 at 13:42

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.