Table 3.

Comparison between Machine Learning and linear regression approaches in stroke patients to predict outcome at discharge.

Stroke Patients
Authors	Algorithms	Sample (n°)	Data Type	Outcome	Accuracy Regression vs. ML Models	Best Features Extracted
Rafiei et al. [36]	Enhanced probabilistic neural network General Linear Model	47	Demographical data Stroke-related data Wolf Motor Function Test performance time, fine motor score, gross motor score Touch sensation (Semmes-Weinstein Monofilament Test) Cognitive function (Montreal Cognitive Assessment) Pretreatment daily arm use (MAL).	Multidimensional assessment (Motor Activity Log, Wolf Motor Function Test, Semmes-Weinstein Monofilament Test of touch threshold, and Montreal Cognitive Assessment).	40–51%/85–91%	Sensation Wolf Motor Function Test Gross Motor Score
Scrutinio et al. [37]	Random Forest ADA-Boost and gradient boosting	1207	Demographical data Stroke-related data Functional Independence Measure cognitive and motor Laboratory findings	Death	75.7%/86.1%	Age Severity Time from stroke Functional Independence Measure
Kim et al. [38]	Deep Neural Network Random Forest	1056	Age Type of stroke Medical Research Council scale scores Modified Brunnstrom classification score Functional Ambulation Category score Presence of motor evoked potentials	Modified Brunnstrom classification and Functional Ambulation Category	84.9%/90% (Deep Neural Network 90%),87–91% (Random Forest)	Presence of motor evoked potentials
Iosa et al. [39]	Artificial Neural Network	2522	Demographical data Stroke related data Bamford Classification Clinical assessment of deficits	Barthel Index	76.6%/74%	Global aphasia Age Neglect
Iosa et al. [28]	Artificial Neural Network	33	Spatio-temporal gait parameters Trunk kinematic parameters during walking	Return to Work	81.3%/93.9%	Double support phase Trunk rotation range
Imura et al. [40]	Support Vector Machine k-Nearest Neighbors Random Forest Decision Tree	481	Demographic data Stroke related data Brunnstrom recovery stage Functional independence measure scores	Home discharge	79.9%/84.0% (k-Nearest Neighbors), 82.6% (Support Vector Machine), 79.9% (Decision Tree), 79.9% (Latent Dirichlet Allocation), 81.9% (Random Forest)	N.R.

Legend: GCS = Glasgow Coma Scale (GCS), Coma Recovery Scale-revised (CRS-r), Glasgow Outcome Scale-Extended (GOS-e), Early Rehabilitation Barthel Index (ERBI), CT = Computed Tomography, N.R. = no reported.