10 Jul · BCV

Meetings

Link to page

Work done:

Balancing Data: Implemented using undersampling, class weighting, and data augmentation.

Training Models: Trained using EEGNet, EEG Transformer and LSTM.

Training Approaches: Conducted within-subject and cross-subject training.

Questions:

Impact of Band-Pass Filtering on Signal Shape:

Does applying a band-pass filter alter the shape of the EEG signal?

⁠

Using MNE package:

Creating RawArray with float64 data, n_channels=16, n_times=7978 Range : 0 ... 7977 = 0.000 ... 63.816 secs Ready. Filtering raw data in 1 contiguous segment Setting up band-pass filter from 1 - 40 Hz

FIR filter parameters --------------------- Designing a one-pass, zero-phase, non-causal bandpass filter: - Windowed time-domain design (firwin) method - Hamming window with 0.0194 passband ripple and 53 dB stopband attenuation - Lower passband edge: 1.00 - Lower transition bandwidth: 1.00 Hz (-6 dB cutoff frequency: 0.50 Hz) - Upper passband edge: 40.00 Hz - Upper transition bandwidth: 10.00 Hz (-6 dB cutoff frequency: 45.00 Hz) - Filter length: 413 samples (3.304 s)

Handling Spikes in EEG Data:

There are significant spikes in the EEG data, and one recommendation is to use EEGNet to address these. What are your thoughts on this approach?

Standardization of Training Data:

It is suggested to standardize the data by subtracting the mean and dividing by the unit variance for each channel (column) before training. This process changes the values, but does it affect the meaning of the channels?

Within-Subject and Cross-Subject Training Approaches:

For within-subject training, I balanced the data by randomly sampling from the maximum classes, sorted the data in a time-based manner, and then split it into two halves (80% training and 20% testing).

For cross-subject training, I combined the data and balanced it without considering the time sequence.

Are these approaches reasonable?