I have an eeg data about 10 minutes. I want to extract some features (e.g. statistical features, power spectrum features, interchannel features, . . .) from this data to apply them to the machine learning algorithms. So I used 3 seconds window with 50% overlap for sliding on the eeg signal. After that because of different ranges of values, I normalized the features with z-score normalization as shown below:
value is feature vector and mu and sigma are the mean and standard deviation of that vector. For further use I want to save these parameters (mu, sigma) to normalize the new signals (test data) with the same parameters before applying to the machine learning algorithms, but in this case I would have too many parameters that saved for each window. What should I do now?
A possible way is that first extract the features from all of the signal (10 seconds) which is time consuming work and then normalize them and save their mu and sigma parameters (for each feature). After that use windowing and also extract the features for windows and normalize them with the saved parameters. This step is for training. Saved mu and sigma from all of the signal also can be used for test data, but I'm not sure that this solution is correct !!!