Table 7 Comparison between the proposed methods and other methods

This work

EEG data analyses

Time and frequency domain features

73.67

(accuracy)

[42]

IMU placed on hand and analyzing data using deep neural network

Automatically learn features about data

97

(accuracy)

[43]

Wrist-worn 3D accelerometers and Deep learning: Convolutional neural networks

Non-negative factorization of frequency features

95

[31]

Bi-axial gyroscope data analyzed by adaptive Kalman filter and a wavelet transform

Spectral-temporal features

95.63

[44]

Features extracted from LFP from the subthalamic nucleus

Power in frequency bands

78

(accuracy)

[15]

Inertial sensors (accelerometer and gyroscope) attached to the index finger and wrist and SVM classifier

Root mean square, average peak power, standard deviation

88.9

(accuracy)

[45]

Accelerometer and gyroscope and bagged ensemble of decision trees

 Sum of absolute differences and sums of squared magnitudes of accelerometer data

82

[17]

Surface electromyogram and acceleration signals

Power at peak frequency, energy of selected wavelet coefficients, Shannon entropy, recurrence quantification parameters

80.2

(accuracy)

[46]

Accelerometers and surface electromyography placed on forearm and shank and dynamic neural network algorithms

Evolving temporal characteristics (energy, autocorrelation)

94.9 (sensitivity)

97.1 (specificity)

[21]

Local field potentials obtained from a DBS system

Energy, variance, zero crossing rate, autocorrelation, information theory, power spectral density magnitudes

86

(accuracy)

[22]

Radial basis function neural network and particle swarm optimization technique based on local field potentials

 Frequency changes between pre-tremor and tremor conditions

89.91

(accuracy)

[16]

Miniature gyroscopes placed on forearm 

Hilbert transform and instantaneous frequency

99.5 (sensitivity)

94.2 (specificity)