Classification of EEG data

I have data of 19x542500 for each subject. Where 19 are channels and 542500 are data points. There are 14 subject for each class.
What i am confused, should i calculate feature on a vector lenght of 542500 , or should i split them into segments.
If segment them, should i calculate feature for each segment.
or
should i average these segments and calculate features

Hi Talha. There are many features of EEG, so the right answer will depend on the feature you are focusing on.

But in general, yes it probably makes sense to segment or downsample the data, either before or after feature extraction.

i believe my question is not clear. I am asking should i convert data to epochs before finding features. Here by segmentation i mean epochs. My data does not have any event. So i am confused should i convert it to epochs or not

Yup, segmentation is the same thing as epoching. Basically just cutting the data from 2D into 3D, where the third dimension is epochs/trials/segments. Typical epoch lengths in the resting-state EEG literature are ~2 seconds. This usually helps minimize the effects of non-stationarities, because brain data tend to be quasi-stationary for short periods of time (hundreds of ms to a few seconds) and highly non-stationary over longer periods of time.

After converting 2D resting stage data to 3D data, do i need to take average to convert it back to 2D. or should i find feature(mean, peak to peak, entropies, variance etc) for 3D data?

Well, there are many many ways to analyze EEG data, so I couldn’t really tell you what you should do. But it often makes sense to extract the features from each segment, and then either average the results together across segments, or use them as repetitions in a feature X observations data matrix.