“…A large number of feature sets have been proposed for PD detection. The vast majority rely on time domain features (such as the mean, range, or cross-correlation) [ 13 , 14 ], frequency domain features (such as the dominant frequency, energy content in a particular band, or signal entropy) [ 15 , 16 ], or a combination of the two [ 17 , 18 , 19 , 20 ]. Some authors have shown that features that are traditionally used for speech processing (e.g., Mel frequency, Cepstral coefficients) are also effective for classifying human motion from accelerometer data [ 21 , 22 ].…”