TABLE 6.

Machine learning techniques; to get patterns and predict intentions from the user for lower-limb prosthetic control.

	ML techniques
Author	Feature extraction	Feature selection	Classifier	Validation technique	Input data	Reported accuracy
Toledo-Pérez et al. (2019)	—	Principal component analysis (PCA)	Support vector machine (SVM)	—	EMG	95%–100%
Su et al. (2019)	—	—	Convolutional neural network (CNN)	Cross-validation	Motion intention from IMUs	94.15% for the able-bodied and 89.23% for amputees
Gupta and Agarwal. (2019)	—	ANOVA, ρ-value <0.05	SVM, linear discriminant analysis (LDA), and neural network (NN)	10-fold cross validation	EMG	96.83 ± 0.28%, 97.45 ± 0.32%, and 97.61 ± 0.22% respectively
Sattar et al. (2021)	Signal mean (SM), variance (σ), skewness (SSK), kurtosis (SK), slope (SS), waveform length (WL), mean absolute value (MAV), root mean square (RMS), Willison amplitude (WAMP), and Zero Crossing (ZC)	—	k-nearest neighbours (KNN)	10-fold cross validation	EMG	Offline 95.8% and 68.1%, real-time 91.9% and 60.1% for healthy and amputated subjects, respectively
Brantley. (2019)	Power spectral density (PSD) and total power in the θ, α, β, and γ bands	PCA	SVM	10-fold cross validation	EEG	80% in offline decoding
Wang et al. (2022)	—	—	Cascade classifiers with a gait phase dependence	—	EMG and IMU	99.13% and 99.39% for the standing and swing phases, respectively
Meng et al. (2021)	RMS	—	SVM	5-fold cross validation	EMG and linear acceleration data of lower limb	98%