MYOELECTRIC control relies on the use of the electromyogram (EMG) signals from the remnant muscles in the amputee s stump as a source of neural control infor- mation. Processing and pattern recognition (PR) of the EMG signals have been at the core of prosthetic control research in the last decade [1] [3]. Previous research has shown that the success of the EMG PR systems mainly depends on the quality of the extracted features, as they are of direct impact on clinical acceptance [4], [5]. In fact, it has been shown that the choice of feature set has a stronger influence on the classification accuracy than the choice of classifier [6], [7]. The main assumption of those systems is that the patterns of recorded EMG signals vary among different motions and have some similarity and repeatability for the same motion [8]. In this direction, many advanced signal processing algorithms have been developed to increase the amount of information that can be extracted from each channel of EMG activity. In the direction of EMG feature extraction, Hudgins set of time domain features [8], which comprises the mean absolute value (MAV), waveform length (WL), slope sign changes (SSC), and number of zero crossings (ZC), have extensively been applied for the classification of hand motions [7]. A number of different feature extraction tech- niques have also been proposed to tackle the more com- plex, multi-channel systems in which Hudgins features strug- gled with [9]. Some of these methods utilized fast Fourier transform (FFT) [10], wavelets [11], wavelet packet trans- form (WPT) [12], cepstral coefficients (CC), Willison ampli- tude (WAMP) [4], sample entropy (SampEnt) [13], reduced spectral moments (RMOM) [14], cardinality of EMG [15], EMG synergies by matrix factorization analysis [16], and autoregressive (AR) model parameters [5] to extract more powerful features. Generally, three-types of features, namely time domain (TD), frequency domain, and time frequency domain features, are dominant in the literature with the TD features being widely adopted for their simplicity and effec- tiveness. A recent study by Phinyomark et al. [17] compared 50 feature extraction methods for EMG pattern recognition and found that the combination of SampEnt, fourth order CC, root mean square (RMS), and WL is the best robust feature subset when classifying EMG data collected over multiple days