Suppose I have collected the hemodynamic responses of participants when they were performing cognitive tasks (e.g. n-back) using a 16-channel functional near-infrared spectroscopy device. I would like to extract some features of the data and perform machine learning to classify participants (e.g. demented vs. not demented / ASD vs. non-ASD).

Can the oxygenated hemoglobin data in all 16 channels of each participant treated as 16 distinct observations? That is, if I have 3 participants, the sample size = 3 participants x 16 channels = 48. I wonder if that is possible? (the data in each channel may correlate with each others)

And, if I use this way to increase the sample size, will it violate any assumptions of machine learning?

  • $\begingroup$ Your data sound photograph-like (or x-ray). Is that fair to say? $\endgroup$ – Dave Feb 8 at 11:24
  • $\begingroup$ The data is in numerical instead of a figure/photograph. All the channels were placed on the forehead and it measured the oxy-hemoglobin in these 16 channels, so each participant got 16 data points. $\endgroup$ – TLL Feb 8 at 14:29
  • $\begingroup$ What is your concern about using the $16$ variables? This seems like a standard regression-type problem where you make a prediction based on measurements you take, whether they are your $16$ variables or the pixels in a photograph. Do you mean as $16$ distinct observations so that $3$ patients means $n=48$? $\endgroup$ – Dave Feb 8 at 16:04
  • $\begingroup$ Sorry for the confusion. What I want to ask is what you mentioned: Can the data in the 16 channels treated as 16 distinct observations so that 3 patients means n=48? $\endgroup$ – TLL Feb 9 at 2:05

Your Answer

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

Browse other questions tagged or ask your own question.