README.md

meg analysis algorithm

algo1 MATLAB

load data using fieldtrip toolbox by function ft_preprocessing select meg channels (name starts with AG*) divided 300s data into 10s epoch/trial label for each category (0-Control, 1-MCI, 2-Dementia) saved into structure s with object data and label into mat files

algo2 python

data loading

load data using scipy.io.mat resampling at 250 Hz for both site A and site B arranging data in format (n_trails, n_channels, n_samples)

filtering in Delta, Theta, Alpha continuous band (0.2-13 Hz)

5th order butterworh bandpass filter save filtered and unfiltered data for train and test plot spectrum

splitting

dividing data into train and validation

Feature extraction , selection

Covariance matrix with 'lwf' estimator Electrode selection for varying values (nelec = 20,22,24,26....40) Vectorization using upper traingular Covariance matrix

MLP and SVM classifier

fitting train data and labels and predicting train, validation and test labels calculating accuracy with confusion matrix

Optimization

based on training and validation accuracy score selecting best prediction

Results

We found best predictions for our model with 96% training accuracy and 91% validation accuracy for MLP classifier for nelec = 28