It sounds like you might be analyzing the EEG data from at least two perspectives
- sub-samples of the EEG, I.e. epochs, for each question
- average overall activation across the entire recording
The individual question epochs could be labeled as "correct" or "incorrect", based on recorded response, and used as part of a supervised machine learning analysis.
The "average overall activation patterns" could be used in, for example, unsupervised cluster analysis or a regression based on percentage of correct responses.
Those are just some ideas based on my very limited experience, so take them with a grain of salt.
I highly recommend the lecture series Introduction to Modern Brain-Computer Interface Design as a reference/overview of many related concepts, such as signal processing and machine learning.