Table 2.

Classification results.

Classification criteria	Linear neural network	SVM	Random forest	XGBoost	Logistic regression
PD/HC 2 classes: HC, PD
Accuracy	0.809 (0.138)	0.842 (0.126)	0.817 (0.099)	0.878 (0.116)	0.840 (0.131)
Precision	0.826 (0.136)	0.862 (0.111)	0.818 (0.107)	0.853 (0.152)	0.856 (0.126)
Recall	0.802 (0.142)	0.839 (0.143)	0.816 (0.104)	0.860 (0.142)	0.824 (0.158)
F1 score	0.792 (0.141)	0.829 (0.136)	0.801 (0.106)	0.841 (0.150)	0.817 (0.151)
H&Y 6 classes: HC, stage 1–5
Accuracy	0.575 (0.177)	0.689 (0.168)	0.682 (0.151)	0.715 (0.140)	0.658 (0.162)
Precision	0.487 (0.241)	0.371 (0.217)	0.423 (0.214)	0.467 (0.233)	0.477 (0.237)
Recall	0.471 (0.206)	0.446 (0.199)	0.464 (0.237)	0.520 (0.236)	0.520 (0.209)
F1 score	0.454 (0.208)	0.388 (0.203)	0.427 (0.218)	0.473 (0.224)	0.472 (0.215)
MDS-UPDRS Part III 4 classes: HC, mild, moderate, severe
Accuracy	0.560 (0.168)	0.636 (0.170)	0.476 (0.144)	0.570 (0.165)	0.628 (0.171)
Precision	0.521 (0.194)	0.579 (0.217)	0.460 (0.177)	0.528 (0.191)	0.557 (0.218)
Recall	0.506 (0.188)	0.613 (0.212)	0.494 (0.149)	0.539 (0.183)	0.600 (0.191)
F1 score	0.477 (0.176)	0.555 (0.206)	0.438 (0.143)	0.501 (0.174)	0.544 (0.198)

^aBold font suggests the best results under a certain classification and evaluation criterion.