Detecting the symptoms of Parkinson’s disease with non-standard video

Table 5 Features selected for the SVM classifier

Type	Joint	Signal	Filter width	Additional detail
Fourier coefficient	Shoulder	Position	0.0	0.47–0.52 Hz
	Shoulder	Position	0.0	3.93–3.97 Hz
	Shoulder	Position	0.0	4.19–4.23 Hz
	Shoulder	Velocity	0.0	0.15–0.21 Hz
Pause metric	Shoulder	Velocity	0.1	\(T=0.60\)
	Elbow	Angular acceleration	0.1	\(T=0.65\)
Regression coefficent	Elbow	Angle	0.1	Slope of amplitude vs. time
	Finger	Position	0.1	Intercept of frequency vs. time
tsfresh	Wrist	Position	0.0	Mean
	Elbow	Velocity	0.0	Longest strike above mean
	Shoulder	Acceleration	0.0	Aggregate linear trend
	Shoulder	Velocity	0.0	First position of maximum
	Elbow	Velocity	0.0	Sum of reccurring values

After optimizing our models as outlined in “Modeling” section, 13 features were selected for training a binary SVM classifier. The feature type, count, associated joint, signal, filter width, and additional details are all presented. Note that a filter width of 0.0 is simply the unfiltered condition.

ISSN: 1743-0003