search

brainhack-school2020 / Alex-A14_Brainhack2020

Creative Commons Zero v1.0 Universal
1 stars 2 forks source link

Sample Size for ML #1

Open Alex-A14 opened 4 years ago

Alex-A14 commented 4 years ago

Hi @vborghe do you think a sample size of 21 is too small to do any meaningful machine learning analyses across participants? In the music data I'm looking at now, the only behavioural data are affective response to music, and a music preferences questionnaire.

The affective response was recorded continuously as participants listened to pieces of music in the scanner (i.e. they indicated how they were feeling by moving a joystick). So there's more data for this.

For the music preferences questionnaire, participants rate their preference for variance genres of music. So if I were to focus on one specific genre, I would only have one rating for each participant (similar to if I was predicting age I guess). Is 21 too few subjects for this? Thanks for any input!

vborghe commented 4 years ago

Hi! I'd say it's definitely on the low end, but the devil is in the details. Correct me if I misunderstood, but you aim at predicting music preference (labels, y, n=21) based on [subjects (sample) x fMRI data (features) = 21 x ?], right? I am not sure I would buy it in a publication, but I think as an exercise in applying ML to fMRI it would work. What are you planning to use in terms of neural features, also meaning, what would be the size of your X?

vborghe commented 4 years ago

I've only glanced at the dataset but it seems like there might enough data within-participant to have fun with... As you mention, the fact that the affective response was recorded continuously is very intriguing!