Is you decom result is for alpha beta ,theta...

i.r mean alpha, mean beta and so.

feature parameter can vary for different study. based on the experimental hypothesis it has to selected

most of the lit paper will explain the same. if not let me know your objective or aim of th work, whats you final goal after feature extraction

Study the scientific publications of Alan S. Gevins. He pretty much defined what is useful and what is not.

Thanks for your answers.

1. Srinivasan Jayaraman, I wanted to study the classify the all the subjects into 2 groups meditators and non meditators by extracting these features.

2.Thank you james.

If I understood your description correctly, you are not analyzing the data in frequency space, which you probably want to do. So, do a FFT transformation and see how much activity there are on different frequency bands. For example, one would assume different levels of alpha activity between those two conditions.

So to put it simply: after artifact removal, you would calculate the powerspectrum for each channel and use those as the feature vector ( so you would have 6 features for each channel: the delta, theta, alpha, beta, gamma, mu as described for example in wikipedia: http://en.wikipedia.org/wiki/Electroencephalography)

I would agree with Ilkka. You should see some pretty large shifts in alpha/beta power as people shift to a meditative state. This shift would be even greater (in theory) in individuals with expertise in meditation. You could then feed in the power spectrum at each site into some sort of classifier (SVD maybe?) which could then be tested on another set of subjects or using a leave one out methodology.

See

supermathematica.com; supermatematica.ro; supermatematicaonline.blogspot.ro

As mentioned by others, spectral decomposition of the signal into frequency regions is a standard technique for quantifying EEG signal content. Do note that your choice of 1-second long windows influences the accuracy of spectral decomposition, especially for low-frequency delta waves. The window is too short to adequately calculate the spectral power of delta waves, as the wavelength of those waves is larger than the actual window. Ideally you would want your window to be (much) larger than the wavelength of the lowest-frequency wave you are trying to quantify, which in general means using windows of at least 4 or even 8 seconds.

Also, what is your reason for using a classification approach "to quantify the changes that occur in ones brain signals when one meditates"? Do you want to be able to determine, on the basis of someone's EEG signal, whether he or she is meditating? In this case, this is a classification problem and your approach is appropriate. On the other hand if you want to determine whether meditation significantly alters EEG signal content, a statistical analysis based of your measurement data is probably recommendable. Depending on your actual study design, (repeated) ANOVA or a linear mixed model analysis could be interesting.

Hello Subodh

Your question is certainly relevant.  The detection of changes in the brain from EEG measurements is a very hot topic these days, with many applications --  brain-computer interfaces are perhaps the most pertinent.  Our own research program is very much engaged in this activity.

We have used a machine learning approach to address this problem, with at least some success.  We have used various feature selection algorithms that identify features that are most relevant to the discrimination between the two states you are trying to distinguish.  Then we feed these selected features into a classifier, which is trained to discriminate between the two classes.  The features can be spectral coherence parameters, power spectral density measurements, wavelet coefficients, etc., etc.

Here are some of our references that explain our approach:

A. Khodayari-Rostamabad, J.P. Reilly, G.M. Hasey, H. de Bruin and D.J. MacCrimmon, A Machine Learning Approach Using EEG Data to Predict Response to SSRI Treatment for Major Depressive Disorder”, Clinical Neurophysiology, Vol. 124, No. 10, Oct. 2013, pp. 1975 - 1985.

Ahmad Khodayari-Rostamabad, Gary M. Hasey, Duncan J. MacCrimmon, James P. Reilly and Hubert de Bruin, “Machine Learning Methodologies Using Pre-Treatment Electroencephalography Can Predict The Symptomatic Response To Clozapine Therapy, Clinical Neurophysiology, Vol. 121, Issue 12, Dec. 2010, pp. 1998 – 2006.

Hope this helps.

Jim Reilly,

Dept. of Electrical and Computer Engineering

McMaster University

Dear Subodh, I'm completely agree with Ilkaa Kosunen, from Aalto University, which suggests you to work in frequency domain. Don't forget to use Hamming windowing in order to reduce spectral leakage and observe carefully the gamma and mu frequency content where you may find some interesting results: something correlated with emotion and motivationin  lies in those bands. Probably meditation interferes with it, isn'it?

Our experience suggests that cross-power spectral density measures (which are the Fourier transform of the cross-correlations between the signals received from two distinct electrodes) are much more indicative of brain state than just power spectral density measure derived from a single electrode.