Table 1.

Overview of studies on commonly used balance/gait metrics and different analyzing techniques

Study	Research Objective	Used Features	Subjects	Measuring equipment	Analysis Method/tool	Study outcome	Max Acc
(Balestrncci et al., 2017)	To investigate the effect of dynamic visual cues on postural control	CoP sway parameters (e.g., M/L and A/P Standard deviation, sway area, etc.)	44 healthy individuals	Force-plate	Statistical Analysis (e.g., t-test, ANOVA)	Gravity-congruent visual motion makes significantly reduced postural sway compared to gravity-incongruent one	N/A
(Fukuchi et al., 2017)	To examine the effect of running speed on gait-biomechanics variables	Kinetics and kinematics (e.g., cadence, stride length, joints angles, joints torque etc.)	28 regular runners	3D motion-capture system & an instrumented treadmill	Statistical Analysis (e.g. one-way ANOVA, Kruskal-Wallis)	Most of gait-biomechanics variables (other than foot-strike) are affected by running speed	N/A
(dos Santos et al., 2017)	To provide the subjects’ full-body 3D kinematics and the GRFs in static balance, while changing support surface and visual cues	Human body’s GRFs, CoPs, and 3D Kinematics (e.g., joints angles)	27 young and 22 older individuals	3D motion-capture system & force platform	Visual3D software and Python programming	Biomechanical characteristics were visualized and modeled (e.g., CoP & COG displacement at the A/P & M/L directions versus time, etc.)	N/A
(Giovanini et al., 2018)	To distinguish between different age groups using force-plate signals with different time-series duration	CoP’s temporal, spectral and spatial features (e.g., root mean square (RMS) distance, sway path, mean frequency, etc.)	older adults (dos Santos & Duarte, 2016) healthy and post-stroke adult (Giovanini et al., 2018)	Force-plate	Statistical Analysis (e.g., Wilcoxon test, MannWhitney U-Test, etc.), Machine Learning (ML) (e.g., K-NN, SVM, MLP, RF, etc.)	For statistical analysis: optimal CoP duration varies based on group under study, For ML analysis: 60 s duration CoP signals is more discriminative	64.9% (RF)
(Reilly, 2019)	To classify fall-risk in older adults using effective feature selection methods (e.g., ReliefF, SAFE, etc.)	Time-domain and Frequency domain features extracted from CoP and Force signals	163 young and older individuals (dos Santos & Duarte, 2016)	Force-plate	Machine Learning (e.g., SVM, K-NN, MLP, NB)	Feature selection methods identified the relevant features successfully, but were incapable of improving classifiers reliability while using only static balance measures	80% (MLP)
(Montesinos et al., 2018)	To differentiate between fallers from non-fallers	CoP’s Approximate entropy (ApEn) and sample entropy (SampEn) with different input parameters	163 young and older individuals (dos Santos & Duarte, 2016)	Force-plate	Statistical Analysis (e.g., threeway ANOVA)	SampEn represents a better choice for the analysis of CoP time-series and to distinguish between groups	N/A
(Cetin & Bilgin, 2019)	To discriminate between young and aged groups	CoP and Forces’ Standard Deviation (STD)	163 young and older individuals (dos Santos & Duarte, 2016)	Force-plate	Machine Learning (e.g., SVM, K-NN, DT, LDA, etc.)	Force signals are better predictors than CoPs for group differentiation while using force-plate measures	81.67% (SVM)
(Ren et al., 2020)	Using AI to evaluate different types of balance control subsystems determined by Mini-BESTest	CoP’s extracted Traditional features (e.g., Mean of CoP displacement, etc.) and pixel-based features (e.g., skewness of gray levels for all pixels)	163 young and older individuals (dos Santos & Duarte, 2016)	Force-plate	Machine Learning Regressors (e.g., RF, MLP, LR, etc.)	Low mean absolute errors (MAE) showed reliability of AI techniques for assessing balance control subsystems	N/A