From: Parkinson’s disease tremor prediction using EEG data analysis-A preliminary and feasibility study
Feature name | Category | Description |
---|---|---|
Entropy(E) | Time-domain | Shannon entropy for wavelet coefficients in each decomposition node of wavelet packet. The final entropy is calculated by the summation of entropy values for all nodes of wavelet packet. \(E=\sum _{i}E\left({s}_{i}\right)\) Where si is the vector of wavelet coefficients in i-th node and E is the entropy. \(E\left({s}_{i}\right)={{s}_{i}}^{2}log{{s}_{i}}^{2}\) |
L-moments (L-scale, L-skewness, L-kurtosis) | Time-domain | L-moment are statistics calculated by linear combination of conventional moments [35]. L-moments are more robust against outliers compared with conventional moments. |
Form Factor (FF) | Time-domain | The ratio of the mobility of the first derivative of the signal to the mobility of the signal [36], where mobility is the ratio of standard deviation for first derivative of time-series and the time-series itself: \(FF=\frac{{\sigma }_{{X}^{{\prime }{\prime }}}/{\sigma }_{{X}^{{\prime }}}}{{\sigma }_{{X}^{{\prime }}}/{\sigma }_{X}}\) |
Sample Entropy (Sen) | Time-domain | Negative logarithm of conditional probability of the successive segmented time-series samples. It is an indicator of time-series complexity [37] |
Root mean square (RMS) | Time-domain | \({X}_{RMS}=\sqrt{\frac{1}{N}\sum _{n=1}^{N}{\left|{X}_{n}\right|}^{2}}\) |
Conventional statistics (median, mean, variance, skewness, kurtosis, higher order statistics (5th and 6th momentums) | Time-domain | The mean value, median value, variance, kurtosis, skewness and 5th and 6th order statistics. These statistics are dependent to the distribution of data points of time-series. |
Peak frequency (Hz) | Frequency-domain | The frequency in which maximum value of power spectral density was observed. |
Band power (Hz) | Frequency-domain | The average power in the input signal. |
Power band width (Hz) | frequency-domain | 3dB bandwidth (half power bandwidth). Using the peridogram power spectrum estimate by a rectangular window, the frequency difference between points in which the spectrum is at least 3dB lower than the maximum point of spectrum. |