The dataset consists of Eeg electrode power features in all power bands(alpha, beta, delta..both relative and absolute) and source power features (obtained after source estimation) in addition to connectivity strengths between the different sources (brain regions). There are as many as 20k features in all.

If i have to predict disease(binary classifier) based on all above features, what approach will yield best accuracy on test sets? I am wondering if i must fit seperate classifiers for each type of feature set (after dimension reduction) and then use the output probabilities obtained to write a meta classifier on top to predict the final disease state.

I am looking for a good discussion on possible approaches and/or a sample solution. Thank you.

  • 1
    $\begingroup$ any dimensionality reduction techniques like PCA ? as u mentioned there are 20k features? $\endgroup$ – venkat krishnan Jul 23 at 8:51
  • $\begingroup$ Sure,what i am specifically interested in is how to choose and to a good extent, a working example of such an ensemble or meta classifier when the features involved are different from one another.Although in this case, i think that the features are correlated because source power estimation is obtained from scalp level electrode power features. $\endgroup$ – Mvkt Jul 23 at 9:35
  • 1
    $\begingroup$ You can actually do it, or try random forest (which is an ensemble itself) and will pick selected features (hopefully the correlated or same sector ones) $\endgroup$ – venkat krishnan Jul 23 at 9:38
  • 2
    $\begingroup$ You've stated this as a classification problem but it may be better suited for prediction. See this. $\endgroup$ – Frank Harrell Jul 23 at 11:14
  • $\begingroup$ @Frank Harrell thank you.sure,i understand.but,even if that is so,i am still unsure how to design the initial models for dimensional reduction and subsequent prediction on unseen data.for ex,i used KernelPCA on the entire datasets and trained using various estimators.i do not get an accuracy of more than 74% on test data.given,most features are correlated sonce the source estimated are obtained by sLoreta analysis of eeg electrode power measures,my concern is whether this dataset can produce higher accuracies if at all.if there is hope,then,what are effective dimension reduction techniques. $\endgroup$ – Mvkt Jul 23 at 12:14

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.