Skip to main content

Table 2 Summary of the features extracted from PSG recordings

From: Feature ranking and rank aggregation for automatic sleep stage classification: a comparative study

Signal Category Feature name
EEG Time domain (F1 to F12) Statistical features (minimum value, maximum value, arithmetic mean, standard deviation, variance, skewness, kurtosis, median), zero-crossing rate, Hjorth parameters (activity, mobility and complexity) [32]
Time–frequency domain (F13 to F26) Features extracted from wavelet packet coefficients including energy of α, δ, β1, β2, θ and spindle bands, total energy of all bands, energy ratio of (\( \frac{\alpha }{\delta + \theta } \), \( \frac{\delta }{\alpha + \theta } \), \( \frac{\theta }{\alpha + \delta } \), \( \frac{\delta }{\theta } \), \( \frac{\alpha }{\theta } \)), statistical features (mean and standard deviation of coefficients in all of the bands)
Entropy (F27 to F30) Spectral entropy, Rényi entropy, approximate entropy, permutation entropy [32]
Non-linear (F31 to F36) Petrosian fractal dimension, teager energy, energy, mean curve length, hurst exponent [32], ISD
EOG Time domain (F37 to F41) Mean, maximum, standard deviation, skewness, kurtosis [58]
Non-linear (F42) Energy [58]
EMG Frequency domain (F43 to F46) Total power in the EMG frequency spectrum, statistical features of EMG frequency spectrum (maximum, mean, standard deviation) [58]
Non-linear (F47 to F49) Energy, ratio of the EMG Signal energy for the current epoch and previous epoch, ratio of the EMG signal energy for the current epoch and next epoch [58]