beapdk
commented
4 years ago Hi Isabelle,
By default, the sklearn library for knn uses uniform weights (all points in each neighborhood are weighted equally), which we did not change. There isn't a lot of litterature regarding the use of EEG to identify subtypes for at-rest data, especially among adults. Some studies that have taken interest in the subject use tasks and there's been mention of underactivated state in the EEG with subtype-specific differences, usually for theta (and alpha) activity and the theta/beta ratio in subgroups of children with ADHD. Therefore, we chose those frequency bands (with the addition of Delta) as key to our model for the eeg dataset exploration. As there's a lack of evidence we though it was appropriate to refer to exploration methods for our data analyses. We used PCA, which is an unsupervised approach to classify features and remedy dimensionality issues, and used these new features for KNN classification (which was accessible and very well documented for beginners in machine learning).
I hope this answers your questions! Beatrice
Good evening,
Interesting research question!
I was wondering what parameters you chose for your model and how you adjusted their weight? What are the typical subtypes traits found in neuro-imaging data with traditional statistical analysis? Finally, I was wondering what oriented the choice of the model that you are using? Is there a specific feature in the data that orient toward this model?
Thank you in advance for all the information provided :)
Have a good evening,
Isabelle